Rare-disease genome interpretation remains constrained by variant-centric approaches that underrepresent the cumulative impact of multiple variants within a gene. GenePy addresses this by generating an individual-level, gene-based burden score integrating variant consequence, allele frequency, and zygosity into a single quantitative metric, supporting panel-agnostic gene prioritisation for unresolved monogenic disorders. We present a fully automated, containerised GenePy workflow orchestrated with Nextflow (DSL2) and deployed in the Genomics England (GEL) Research Environment via Lifebit CloudOS, enabling scalable, auditable computation across population-scale cohorts under governance controls.

Aggregated germline whole-genome sequencing VCFs are ingested as per-chromosome chunks and linked to participant metadata within the secure environment. To reduce compute and storage while retaining interpretable regions, variants can be subset to CCDS coding regions with ±25 bp flanks prior to execution. The workflow partitions VCFs into genomic windows for parallel processing and performs variant normalisation, multi-source annotation (Ensembl VEP on GRCh38 gene models, population allele frequencies, and deleteriousness metrics such as CADD), and multi-criteria quality control.

Harmonisation resolves multiallelic representation, enforces consistent mapping between variant records and transcript-level annotations, and merges overlapping chunk boundaries into canonical variant sets; sex-aware handling of chromosome X preserves allele-dosage semantics across karyotypes.

Annotations are aggregated into per-gene metadata and re-aggregated genome-wide to generate a canonical record per gene with harmonised identifiers. GenePy scoring rescales deleteriousness, integrates allele frequency and genotype state, and outputs a gene-by-participant matrix suitable for downstream statistics, phenotype-driven prioritisation, and machine-learning analyses. This portable framework enables reproducible, governance-compliant gene-burden computation for large rare-disease sequencing datasets."

Timothy Syndrome Alliance (TSA) is a UK-based international charity dedicated to improving diagnosis, treatment, and support for those with Timothy Syndrome and CACNA1C-Related Disorders. These conditions are caused by changes in the CACNA1C gene, which codes for a calcium channel. Rare variants in the gene can result in a range of significant health problems spanning cardiac, neurodevelopmental, neurological, and/or physical abnormalities.

Within the rare disease space, TSA have been instrumental in identifying more than 350 patients with rare variants in CACNA1C; providing counselling services and community via patient advocacy boards, family days and support groups. TSA collaborates with researchers worldwide but particularly with the NMHII at Cardiff University.

In partnership with an international panel of CACNA1C experts and the TSA patient community, TSA have recently co-led the development through Delphi consensus of diagnostic guidelines for individuals with CACNA1C variants. The organisation is also developing a global research portal that integrates genetic variants, clinical features, and functional data supported by the CACNA1C Community Registry.

Notable achievements of TSA include award-winning films, the establishment of the CACNA1C patient registry, and recognition from the 2022 Gene People Awards, 2025

Rare Disease Research UK PPIE Awards and more. In 2024, TSA’s collaborative model was further acknowledged through a CZI Rare As One (cycle 3) grant.
By combining trusted community partnerships with real-world evidence, clinical consensus and cross-sector collaboration, the TSA’s work on CACNA1C is opening pathways for earlier diagnosis, trial readiness and sustainable research infrastructure.

Background: Single-base substitution (SBS) mutations are increased owing to unrepaired DNA replication errors in mismatch repair-deficient (MMRd) cancers. Excess CpG>TpG mutations have been reported in MMRd cancers defective in mismatch detection (dMutSα), but not in mismatch correction (dMutLα). Somatic CpG>TpG mutations conventionally result from unrepaired spontaneous deamination of 5’-methylcytosine throughout the cell cycle, causing T:G mismatches and signature SBS1. It has been proposed that MutSα detects those mismatches, prior to repair by base-excision repair (BER). However, other evidence is contradictory to this hypothesis.

Methods: We analysed mutation spectra and COSMIC mutation signatures in whole-genome sequencing data from 1,803 colorectal cancers and 596 endometrial cancers from the 100,000 Genomes Project. We mapped each C>T mutation to its normal DNA methylation state, replication timing, transcription strand, and replication strand, to investigate the mechanism(s) by which these mutations arise.

Results: We confirmed that dMutSα tumours had higher CpG>TpG burdens than dMutLα tumours. We found evidence indicating that the SBS1 excess in dMutSα cancers did not come from 5’-methylcytosine deamination alone. In contrast to BER-deficient tumours, CpG>TpG mutations were biased to the leading DNA replication strand at similar levels in dMutSα and dMutLα cancers, suggesting an origin in DNA replication.

Conclusions: There is a CpG>TpG excess specific to dMutSα tumours, consistent with previous reports. Contrary to some other studies, the similar leading replication strand bias in both dMutSα and dMutLα tumours indicates that at least some of the excess CpG>TpG mutations arise via DNA replication errors, and not primarily via the replication-independent deamination of 5’-methylcytosine.

De novo mutations (DNMs) are a major cause of rare disease, yet the genetic modulators of germline mutagenesis – beyond age at conception or environmental exposures like smoking and chemotherapy – remain poorly defined. To investigate this, we performed a rare-variant association study in ~10,500 whole-genome sequenced parent-offspring trios from the 100,000 Genomes Project, one of the largest resources for DNM research. We tested for association between rare damaging variants in the parents in ~18,400 genes against both the total DNM count in the offspring (SNVs and INDELs), and the number of de novo SNVs derived from the relevant parent (inferred through read-backed phasing). For statistical robustness and association power, we used Regenie and performed a meta-analysis using Stouffer’s method across ancestry groups, while controlling for parental age and technical covariates.

After the Cauchy combination of different tests (SKATO, annotation), we identified four significant associations (FDR<5%). We associated de novo INDEL counts with maternal pLoF/damaging-missense variation in TUBA3D (P_EUR = 6.3E-07; P_meta = 2.1E-08), a gene encoding β-tubulin in which defects could conceivably compromise spindle integrity during oocyte meiosis, increasing indel formation. We also associated increased total DNM counts with paternal nonsynonymous variation in PDCD1 (P_EUR = 1.1E-05; P_meta = 5.1E-07), a gene encoding the PD-1 immune checkpoint. Lastly, APOC1 and PRAMEF15 were significant but may represent technical artifacts (MAC=5). These findings are preliminary and ongoing work involves extending the analysis to additional sequenced trio cohorts to increase statistical power.

Background
Whole genome sequencing (WGS) bioinformatic filtering pipelines are essential in prioritising variants likely to result in genetic diagnoses, streamlining the requirement for manual assessment. Whilst efficient, they can exclude pathogenic variants outside of predetermined parameters, such as intronic, synonymous, structural, and complex variants. Therefore, there exists an “unresolved” patient population who have undergone WGS, have not received a genetic diagnosis, and harbour a causative variant.

Methods
This project sought to find new genetic diagnoses in unresolved patients tested by the WGS service in the South West Genomic Laboratory Hub using a reanalysis approach developed by the Genomics England Diagnostic Discovery team, focusing on selected genes associated with polycystic kidney disease and hereditary spastic paraplegia.

Results
This project reanalysed 319 unresolved cases and identified 12 candidate variants in 17 patients, representing a 5% diagnostic yield. Many variants had strong in silico splicing predictions (SpliceAI). Notably, this project identified a novel heterozygous intronic PKD1 splicing variant, NM_001009944.3(PKD1):c.1606+44T>A, in seven patients from the Northeast of England. The predicted splicing impact of this variant was confirmed by RNA testing. Additional analysis found these patients to share a haplotype, suggesting this to be a founder variant in this geographical region.

Conclusion
This project provides a model workflow for genomic laboratories for translational “Diagnostic Discovery”-type reanalysis of complex variants in specific patient cohorts. In future, applying this approach in other disease groups, particularly those which are monogenic or where few genes are responsible for the disease burden, may allow the discovery of previously missed diagnoses.

Background
Genomic information obtained from next generation sequencing analysis is already entering clinical practice by providing screening and diagnosis for rare disorders.

Objective
The Generation Study (2) run by a partnership of Genomics England and the National Health Service (NHS) continues the evolution of genomic research. This groundbreaking study will sequence the genomes of 100,000 newborns with the objective of identifying affected babies with 200 rare diseases, enabling prompt treatment and improved health outcomes. The results will inform future decisions on using genome sequencing to support newborn screening, and improve our ability to diagnose and treat genetic conditions.

Local approach
BWC has been recruiting participants to the Generation Study since June 2024. As a regional centre it provides specialist care to women and children, with 641,000 visits and 7500 births/year. The BWC Generation Study research team has now developed a streamlined process which incorporates recruitment alongside routine appointments, maximising study participants without an increased work burden for the clinical team, or increasing attendances. BWH is the highest single recruiter outside London, having consented over 2800 participants.

Conclusion
The BWC recruitment approach to the Generation Project demonstrates that implementing genomic screening into the NHS as part of routine care is feasible. Following data analysis and validation, it is hoped that the project results will lead to new diagnoses, personalised management and improved outcomes for newborns and the NHS.

Primary lymphoedema (PL) is a hereditary disorder caused by abnormal lymphatic vessel (LV) development or dysfunction. The endothelial transcription factor ERG is essential for lymphatic endothelial cell (LEC) identity and vessel integrity. We have identified PL patients carrying ERG coding variants that reduce ERG DNA binding and transcriptional activity. Current diagnostic pipelines focus on coding variants, with ~40 genes associated with the disease. However, these explain only 40% of cases and the contribution of non-coding variants to PL remains unknown. We aim to map ERG cis-regulatory elements in LECs and determine whether ERG non-coding variants are causative for PL. Using transcriptomic and epigenomic datasets we identified 54 cis-regulatory elements within a 1.5 Mb ERG topologically associated domain in human dermal LECs. Analysis of 205 PL patient genomes from the UK 100,000 Genomes Project identified 290 rare variants (AF<0.01) within putative ERG enhancers. In silico prioritisation, using tools such as VEP, MotifBreakR and ChromBPNet, identified 31 high confidence candidates. Two variants, rs540276985 and rs35051080, were selected for preliminary functional testing. Luciferase assays confirmed enhancer activity in both regions, with rs540276985 showing allele-specific activity. Motif analysis predicted loss of NFATC1 and GATA2 transcription factor binding at rs540276985; transactivation reporter assays with GATA2 overexpression reduced alternative-allele activity, whilst EMSA demonstrated impaired NFATC1, but not GATA2, binding, suggesting NFATC1 loss underlies reduced enhancer function in this variant. Ongoing functional studies will confirm which of the 31 variants disrupt ERG expression in LEC, supporting a role for pathogenic non-coding variants in causing lymphatic anomalies.

The Generation Study uses newborn whole genome sequencing to detect 200+ rare, treatable genetic conditions that present early in life and (where the majority) are not captured by standard NHS screening.

Implementation at BWCH required structural redesign. Two research delivery posts and a Project Manager funded by Genomics England(GEL) enabled integration despite capacity constraints. Unlike other sites, the Regional Results Coordinator role was embedded within the delivery team for cross-cover, recruitment, clinical support and mapping of eligible condition pathways for safe study opening.

A multi-level workforce strategy replaced 1:1 training, engaging clinical/community teams and interpreter-specific sessions. Internal communication via screensavers and staff networks to increase awareness. A five-minute training video standardised cord sampling. Co-ordination with the baby clinic provided contingency for capillary samples, improving missed sample rates from amber to green.

Recruitment included: community midwife study introduction at 16 weeks; 20-week scans with translated PIS; same-day(1:1) consenting with partners/interpreters; electronic record alerts; glucose clinics; inpatient and elective caesarean waiting areas; digital engagement including BBC. Self-referrals and retrospective consent were enabled.

Cultural and faith-based barriers were addressed through GEL collaboration by approaching Imams and supporting discussions around ""God's-will,"" reassuring families about NHS-funded, treatable conditions. Participant ethnicity(Jan2026): Pakistani-11.4%, Indian-9.1%, Bangladeshi-2.6%, Black African-6.3%, other-4.1%.

Impact: 2,766 total recruits (Feb2026); monthly recruits increased from 27(Jun2024) to 203(Feb2026); birthing volume participation rose from 23%(Aug2025) to 32%(Jan2026).

Workforce sustainability was strengthened via cross-cover, volunteers and recognition through certificates/awards. Embedding the study in routine pathways improved accessibility, inclusion, and provides a scalable research delivery framework.

Generation Study uses newborn whole genome sequencing to detect 200+ rare, treatable genetic conditions that present early in life and (where the majority) are not captured by standard NHS screening.

Implementation at BWCH required structural redesign. Two research delivery posts and a Project Manager funded by Genomics England(GEL) enabled integration despite capacity constraints. Unlike other sites, the Regional Results Coordinator role was embedded within the delivery team for cross-cover, recruitment, clinical support and mapping of eligible condition pathways for safe study opening.

A multi-level workforce strategy replaced 1:1 training, engaging clinical/community teams and interpreter-specific sessions. Internal communication via screensavers and staff networks to increase awareness. A five-minute training video standardised cord sampling.

Co-ordination with the baby clinic provided contingency for capillary samples, improving missed sample rates from amber to green.

Recruitment included: community midwife study introduction at 16 weeks; 20-week scans with translated PIS; same-day(1:1) consenting with partners/interpreters; electronic record alerts; glucose clinics; inpatient and elective caesarean waiting areas; digital engagement including BBC. Self-referrals and retrospective consent were enabled.

Cultural and faith-based barriers were addressed through GEL collaboration by approaching Imams and supporting discussions around ""God's-will,"" reassuring families about NHS-funded, treatable conditions. Participant ethnicity(Jan2026): Pakistani-11.4%, Indian-9.1%, Bangladeshi-2.6%, Black African-6.3%, other-4.1%.

Impact: 2,766 total recruits (Feb2026); monthly recruits increased from 27(Jun2024) to 203(Feb2026); birthing volume participation rose from 23%(Aug2025) to 32%(Jan2026).

Workforce sustainability was strengthened via cross-cover, volunteers and recognition through certificates/awards. Embedding the study in routine pathways improved accessibility, inclusion, and provides a scalable research delivery framework.

Gene2Phenotype (G2P, https://www.ebi.ac.uk/gene2phenotype) exists to improve diagnostic yield in rare Mendelian disease. It openly shares expert-curated, clinician-reviewed gene-disease associations and is updated twice a month with information from the latest publications. The largest gene panel catalogues developmental disorders, but information is also available on cardiac, eye, skeletal and skin disorders and cancer, with obesity and immune panels in development.

Detailed, standardised descriptions of molecular mechanism at disease domain level are needed to facilitate more accurate clinicogenomic diagnosis and therapy development. To help address this, the variant types observed in affected individuals, inferred variant consequences, mechanism classifications and supporting functional evidence are recorded in G2P. This information is available via a RESTful API and the G2P website, which supports intuitive data discovery, browsing and download. G2P is a founder member of the GenCC, the international collaboration collating the strength of gene-disease associations.

We have developed a literature surveillance pipeline to rapidly identify new disease-relevant publications, with regular searches of the entire medical literature. This has direct clinical impact, for example, new evidence has led to gene-disease confidence assertions being upgraded to clinically reportable status.

The G2P Developmental Disorders panel is incorporated into the Paediatric Disorders gene panel (R27) available in the National Genomic Test Directory. To help accelerate genomic diagnosis, we developed a tool based on the Ensembl Variant Effect Predictor to filter genotypes obtained from genomic sequencing. VEP-G2P is used in the Scottish Clinical Exome Service, with knowledge from G2P, to support rapid, robust identification of genotypes for prioritisation. 

Gene2Phenotype (G2P, https://www.ebi.ac.uk/gene2phenotype) exists to improve diagnostic yield in rare Mendelian disease. It openly shares expert-curated, clinician-reviewed gene-disease associations and is updated twice a month with information from the latest publications. The largest gene panel catalogues developmental disorders, but information is also available on cardiac, eye, skeletal and skin disorders and cancer, with obesity and immune panels in development.

Background:
CASK-associated neurodevelopmental disorders present with a broad spectrum of neurological, developmental, behavioural, and clinical features, with a clear genotype-phenotype relationship yet to be established and significant heterogeneity within the disorder. Currently, the existing literature focuses on congenital and neurological phenotypes; less is known about the lived experiences of children and families. This study characterises a UK cohort enrolled in the GenROC observational cross-syndrome study, focusing on lived experience and multisystem involvement.

Material and Methods:
Parents of children with confirmed CASK variants completed structured questionnaires capturing diagnostic experiences, developmental milestones, physical and mental health, and education. Data is represented in descriptive analyses. For each participant, their responsible clinical geneticist completed a clinical questionnaire including variant data.

Results:
Eleven children (10 female, 1 male; 15 months–16 years) were included, including four novel CASK variants. Universal delays in sitting, walking, and speech were reported, with substantial heterogeneity in milestone attainment. Multisystem symptoms were common, including visual impairment, feeding difficulties, altered muscle tone, and structural brain changes, with varied behavioural and mental health issues. All school-aged participants attended specialist educational settings, and families reported significant socioeconomic impact: all experienced work changes due to caregiving needs.

Conclusion:
The GenROC cohort highlights the extensive developmental, medical, behavioural, educational, and financial burden of CASK-related disorders. The cohort included variants across different domains of the protein, which provides valuable in-depth data to better understand the genotype-phenotype relationship. Findings underscore the need for coordinated multidisciplinary care, clearer diagnostic communication, and strengthened financial support systems for affected families.

Background:
The Generation Study (GS) began recruiting participants for genomic newborn screening (gNBS) of >200 treatable childhood conditions in 2024. We aim to assess its impact on healthcare usage.

Methods:
We compared two birth cohorts using linked Hospital Episode Statistics (HES) data: (1) GS children born between April 2024–March 2025 with follow-up to March 2025; (2) English children, i.e. HES-recorded births in NHS hospitals in England between April 2022–March 2023, with follow-up to March 2023. We mapped gNBS conditions to hospital diagnosis codes to flag English children with gNBS-similar conditions. We defined cases as GS children with confirmed gNBS conditions, or English children with gNBS-similar conditions; controls were children without gNBS/gNBS-similar conditions. We calculated planned/unplanned hospital admission rates per 100 person-years by cohort and case/control status.

Results:
Cases comprised 19/4353 (0.4%) of GS children and 3963/548526 (0.7%) of English children. A higher proportion of GS children were non-White, and from West Midlands or London, than English children. Planned/unplanned admission rates (95%CI) were 35.1 (4.2–126.7) and 70.2 (19.1–179.7) for GS cases, and 156.2 (150.1–162.5) and 276.8 (268.7–285.2) for English cases. Corresponding rates were 11.9 (10.0–14.1) and 65.4 (60.8–70.3) for GS controls, and 6.1 (6.0–6.2) and 52.5 (52.2–52.8) for English controls.

Conclusion:

Interim findings showed higher hospitalisation rates for English cases than GS cases, but lower rates for English controls than GS controls, likely due to demographic differences and limited data. Ongoing recruitment will provide greater insight into subgroups and additional outcomes.

Microphthalmia is a rare congenital eye defect characterised by a small, underdeveloped eye. Notably, microphthalmia exhibits genetic heterogeneity, in which mutations in different genes can result in a similar clinical phenotype. Although gene panel testing covering 147 genes is currently used by clinicians in the United Kingdom for the genetic diagnosis of microphthalmia, approximately 70% of patients remain without a molecular diagnosis. This highlights the need to identify additional genes associated with the condition. By examining the de novo variants within the unresolved microphthalmia cohort recorded in Genomics England, we identified an individual with severe microphthalmia who had a heterozygous missense variant in CCNC c.494T>C, p. (Asp165Gly). Further examination of the variant suggests that this change is rare with predicted pathogenicity. To examine the effect of CCNC on eye development in vivo, we generated a ccnc knockout zebrafish model using CRISPR/Cas9. ccnc knockout embryos showed significant reduction in eye diameter, resembling the microphthalmia phenotype observed in the human proband. Together, these results support CCNC as a novel microphthalmia-associated gene, and highlight the value of Genomics England whole-genome sequencing dataset in discovering novel gene candidates in rare genetic diseases.

Introduction
Cancer is the leading cause of death due to disease in children across Europe. Despite improvements in patient outcomes in recent decades, some paediatric solid tumours remain refractory to treatment or relapse following treatment response; some studies estimate a 5-year overall survival as low as 20-30% for these refractory/relapsed tumours. This is underscored by a limited understanding of the T cell microenvironment in paediatric solid tumours and how it could be leveraged for therapeutic intervention.

Methods
To understand the T cell microenvironment in paediatric solid tumours, we are characterising determinants of variability in tumour-infiltrating T cell fractions. Here we apply bioinformatics tools such as ImmuneLENS to tumour and germline whole-genome sequencing from the NHS 100,000 Genomes Project and Genomic Medicine Services, with a particular focus on paediatric CNS tumours and bone tumours.

Results
Bone tumours were generally devoid of high alpha-beta T cell fractions. However, 26.8% of bone tumours had a high gamma-delta T cell fraction; high gamma-delta T cell fraction was associated with distinct copy number changes, homologous recombination deficiency, and shared V/J segment usage. CNS tumour microenvironments were generally composed of alpha-beta T cells with shared V/J segment usage; their T cell fractions negatively correlated with age and tumour mutational burden; hypermutation only caused high T cell fraction in a subset of cases.

Conclusion
The regulation of the T cell microenvironment differed between anatomical compartments. A high fraction of gamma-delta T cells in bone tumours could enable stratification of patients likely to benefit from novel interventions e.g., engineered gamma-delta T cells."

Background
Inhibiting the immune checkpoint proteins programmed cell death protein 1 (PD-1) and lymphocyte activation gene (LAG-3) promotes anti-cancer immune responses. Targeting both PD-1 and LAG-3 has demonstrated success in melanoma treatment. PD-1 inhibitors are also approved for treatment of cancer patient populations based on features such as tumour mutational burden (TMB), suggesting efficacy of these medications across sites. We therefore investigated the repurposing potential of PD-1 and LAG-3 inhibitors in combination for cancer treatment, and whether TMB modifies these effects.

Material and Methods
We selected genetic instruments to proxy plasma PD-1 or LAG-3 protein levels using summary statistics obtained from genome- or exome-wide association studies performed in UK Biobank participants. We also constructed a cohort of Genomics England participants with a primary, malignant cancer diagnosis. Effects of genetically proxied PD-1 or LAG-3 lowering or interactions between these proteins and TMB on risk of cancer-specific mortality were estimated. Hazard ratios (HRs) are reported per SD increase in genetic susceptibility to decreased protein levels.

Results
Preliminary evidence does not support effects of genetically proxied PD-1 or LAG-3 plasma protein levels on risk of mortality (HR=PD-1:1.02 [0.96-1.08]; LAG-3:1.04 [0.98-1.10]; PD-1/LAG-3 interaction:1.04 [0.98-1.11]) or TMB modifying these effects (HR=PD-1/TMB interaction:0.99 [0.93-1.06]; LAG-3/TMB interaction:0.96 [0.90-1.02]; PD-1/LAG-3/TMB interaction:0.99 [0.93-1.06]). Findings were similar for any cancer- or all-cause mortality outcomes

Conclusion
We did not observe evidence to support repurposing of PD-1 or LAG-3 inhibitors as monotherapy or in combination for cancer treatment irrespective of site and patient TMB, however analyses were likely limited by statistical power.

Background
Inhibiting the immune checkpoint proteins programmed cell death protein 1 (PD-1) and lymphocyte activation gene (LAG-3) promotes anti-cancer immune responses. Targeting both PD-1 and LAG-3 has demonstrated success in melanoma treatment. PD-1 inhibitors are also approved for treatment of cancer patient populations based on features such as tumour mutational burden (TMB), suggesting efficacy of these medications across sites. We therefore investigated the repurposing potential of PD-1 and LAG-3 inhibitors in combination for cancer treatment, and whether TMB modifies these effects.

Material and Methods
We selected genetic instruments to proxy plasma PD-1 or LAG-3 protein levels using summary statistics obtained from genome- or exome-wide association studies performed in UK Biobank participants. We also constructed a cohort of Genomics England participants with a primary, malignant cancer diagnosis. Effects of genetically proxied PD-1 or LAG-3 lowering or interactions between these proteins and TMB on risk of cancer-specific mortality were estimated. Hazard ratios (HRs) are reported per SD increase in genetic susceptibility to decreased protein levels.

Results
Preliminary evidence does not support effects of genetically proxied PD-1 or LAG-3 plasma protein levels on risk of mortality (HR=PD-1:1.02 [0.96-1.08]; LAG-3:1.04 [0.98-1.10]; PD-1/LAG-3 interaction:1.04 [0.98-1.11]) or TMB modifying these effects (HR=PD-1/TMB interaction:0.99 [0.93-1.06]; LAG-3/TMB interaction:0.96 [0.90-1.02]; PD-1/LAG-3/TMB interaction:0.99 [0.93-1.06]). Findings were similar for any cancer- or all-cause mortality outcomes.

Conclusion
We did not observe evidence to support repurposing of PD-1 or LAG-3 inhibitors as monotherapy or in combination for cancer treatment irrespective of site and patient TMB, however analyses were likely limited by statistical power.

Background
Cerebral palsy (CP) affects approximately 1 in 400 UK births and is the most common cause of childhood-onset physical disability. While 9–36% of individuals with CP harbour monogenic conditions, whole genome sequencing (WGS) utility within the NHS remains unevaluated and optimising patient selection remains a clinical challenge. This study assessed WGS diagnostic yield, and the performance of Human Phenotype Ontology (HPO)-based machine learning prioritisation across two complementary UK cohorts.

Methods
Eighty-six children with CP were prospectively recruited from five NHS specialist clinics (April 2023–November 2024) and underwent gene-agnostic trio WGS using Emedgene™ for variant prioritisation. Variants were classified following ACGS guidelines and confirmed by NHS Genomic Laboratory Hubs. Using an expanded set of 107 individuals, a supervised Principal Component Analysis followed by Linear Discriminant Analysis (PCA-LDA) model was developed using standardised HPO terms to estimate diagnostic probability and support pre-test triage and validated using 117 individuals from the Genomics England Cerebral Palsy (GELCP) cohort.

Results
In the NHS cohort, pathogenic or likely pathogenic variants were identified in 12.8% of participants, with a further 9.3% carrying variants strongly suggestive of disease causation — approximately one in five children had clinically relevant genetic findings. The HPO-based model demonstrated strong diagnostic discrimination (NHS cohort AUC 0.88; GELCP validation AUC 0.69), with intellectual disability and multisystem phenotypes as positive predictors.

Conclusion
WGS demonstrates meaningful clinical utility in CP across NHS settings. HPO-driven machine learning offers a practical, scalable tool for pre-test counselling and patient selection, warranting validation in larger, diverse cohorts.

Introduction
As use of genomics expands in healthcare, many countries are exploring genomic newborn screening (gNBS) to enhance newborn screening. The Generation Study, led by Genomics England in partnership with NHS England, offers gNBS to 100,000 newborns to screen for more than 200 early onset genetic conditions. Around 1% of babies are expected to receive a condition suspected result from screening and have confirmatory testing.

Aim
To explore parents’ experiences of receiving condition suspected results and clinicians’ experiences of returning these results.
Methods
To date, nine semi-structured interviews have been conducted with 10 parents whose child received a condition suspected result across a range of rare genetic conditions, and 12 interviews with clinicians responsible for returning results from multiple paediatric specialties (including three paired clinician-parent interviews). Interviews were summarised using Rapid Research Evaluation and Appraisal Lab (RREAL) rapid assessment procedure sheets and analysed using framework analysis to identify themes across interviews.

Results
Parents described initial shock and anxiety, followed by gradual adjustment and recognition of the value of early diagnosis. Clinicians generally reported that returning results aligned with their existing experience of delivering genetic diagnoses and didn’t substantially increase their workload. However, returning unexpected results to families with previously well infants required careful preparation and sensitive communication. Both parents and clinicians highlighted gaps in communication, emotional support and information following diagnosis.

Conclusion
These early findings highlight key relational and organisational considerations in returning genomic screening results and provide insights to inform the future implementation of gNBS within NHS newborn screening pathways. "

Background
Cerebral palsy (CP) affects approximately 1 in 400 UK births and is the most common cause of childhood-onset physical disability. While 9–36% of individuals with CP harbour monogenic conditions, whole genome sequencing (WGS) utility within the NHS remains unevaluated and optimising patient selection remains a clinical challenge. This study assessed WGS diagnostic yield, and the performance of Human Phenotype Ontology (HPO)-based machine learning prioritisation across two complementary UK cohorts.

Methods
Eighty-six children with CP were prospectively recruited from five NHS specialist clinics (April 2023–November 2024) and underwent gene-agnostic trio WGS using Emedgene™ for variant prioritisation. Variants were classified following ACGS guidelines and confirmed by NHS Genomic Laboratory Hubs. Using an expanded set of 107 individuals, a supervised Principal Component Analysis followed by Linear Discriminant Analysis (PCA-LDA) model was developed using standardised HPO terms to estimate diagnostic probability and support pre-test triage and validated using 117 individuals from the Genomics England Cerebral Palsy (GELCP) cohort.

Results
In the NHS cohort, pathogenic or likely pathogenic variants were identified in 12.8% of participants, with a further 9.3% carrying variants strongly suggestive of disease causation — approximately one in five children had clinically relevant genetic findings. The HPO-based model demonstrated strong diagnostic discrimination (NHS cohort AUC 0.88; GELCP validation AUC 0.69), with intellectual disability and multisystem phenotypes as positive predictors.

Conclusion
WGS demonstrates meaningful clinical utility in CP across NHS settings. HPO-driven machine learning offers a practical, scalable tool for pre-test counselling and patient selection, warranting validation in larger, diverse cohorts.

Background: Neurodevelopmental disorders (NDDs) frequently co-occur with epilepsy, with many cases driven by rare de novo variants (DNVs). However, up to 50% of individuals remain undiagnosed, suggesting that many disease genes remain undiscovered. Moreover, the biological mechanisms underlying NDD-associated seizures are poorly understood. Here, we assembled a large genomic dataset to discover new epilepsy genes and improve our understanding of epilepsy in NDDs.

Material and Methods: We aggregated and curated DNVs from 15 independent NDD cohorts and applied the mutational model DeNovoWEST to test for gene-level DNV enrichment in protein-coding genes. Candidate epilepsy-associated genes were further evaluated by pathway (IPA) and gene-set enrichment analyses, tissue-specific expression profiling (GTEx and BrainSpan data), and phenotype enrichment analysis.

Results: We analysed 26,238 parent-offspring trios affected by NDDs, of which 5,455 also had epilepsy. Among the 188 genes significantly enriched for DNVs, 178 were established NDD genes. We identified 10 constrained and brain-expressed genes not previously linked to epilepsy. Pathway analyses supported their involvement in plausible neurobiological pathways. ZSCAN5A showed significantly higher phenotypic similarity relative to the broad NDD cohort (Monte Carlo p = 0.044), whereas all other genes demonstrated phenotypic similarity comparable to that of known NDD and epilepsy genes. Finally, we identified gene sets enriched for DNVs in NDDs with epilepsy, including ion channels and neuronal activity.

Conclusion: By integrating de novo variants from large NDD and epilepsy cohorts, we identified 10 novel disease genes, advancing our understanding of epilepsy’s molecular basis in NDDs and potentially enabling new diagnoses for affected individuals."

Background: Neurodevelopmental disorders (NDDs) frequently co-occur with epilepsy, with many cases driven by rare de novo variants (DNVs). However, up to 50% of individuals remain undiagnosed, suggesting that many disease genes remain undiscovered. Moreover, the biological mechanisms underlying NDD-associated seizures are poorly understood. Here, we assembled a large genomic dataset to discover new epilepsy genes and improve our understanding of epilepsy in NDDs.

Material and Methods: We aggregated and curated DNVs from 15 independent NDD cohorts and applied the mutational model DeNovoWEST to test for gene-level DNV enrichment in protein-coding genes. Candidate epilepsy-associated genes were further evaluated by pathway (IPA) and gene-set enrichment analyses, tissue-specific expression profiling (GTEx and BrainSpan data), and phenotype enrichment analysis.

Results: We analysed 2

6,238 parent-offspring trios affected by NDDs, of which 5,455 also had epilepsy. Among the 188 genes significantly enriched for DNVs, 178 were established NDD genes. We identified 10 constrained and brain-expressed genes not previously linked to epilepsy. Pathway analyses supported their involvement in plausible neurobiological pathways. ZSCAN5A showed significantly higher phenotypic similarity relative to the broad NDD cohort (Monte Carlo p = 0.044), whereas all other genes demonstrated phenotypic similarity comparable to that of known NDD and epilepsy genes. Finally, we identified gene sets enriched for DNVs in NDDs with epilepsy, including ion channels and neuronal activity.

Conclusion: By integrating de novo variants from large NDD and epilepsy cohorts, we identified 10 novel disease genes, advancing our understanding of epilepsy’s molecular basis in NDDs and potentially enabling new diagnoses for affected individuals.

Transport of intracellular cargo is driven by nanoscale motors which translocate along cytoskeletal filaments powered by hydrolysis of ATP. Myosin VI (MYO6) is a unique myosin motor that moves towards the minus-end of actin filament and plays a crucial role in endocytosis, mitophagy and actin cytoskeletal organization. While a few pathogenic MYO6 variants have been linked to cardiomyopathy as well as congenital- or late-onset progressive hearing loss, the full spectrum of genotype-phenotype associations remains undefined.

This project aims to classify and experimentally characterise novel MYO6 mutations found within the Genomics England (GEL) 100,000 Genomes Project. 203 variants of MYO6 have been published in PubMed and the Human Gene Mutation Database, which were matched to GEL cohort data containing 65 missense and nonsense mutations. Out of these, 57 GEL variants remain unpublished, with only 5 variants linked to known clinical MYO6 phenotypes. In this project, in silico prioritization was conducted using Exomiser, REVEL, CADD and AlphaMissense scores, with 22 variants scoring “likely pathogenic” in all four scores. These highly-ranked variants are currently being analysed via a newly established pipeline of cell-based assays to measure protein stability, cellular targeting, cargo adapter binding and motor activity. This poster will present the findings of these experiments, aiming to evaluate whether these MYO6 variants have a potential to be disease-causing.

Overall, this project creates a pipeline to establish a link between gene mutations and MYO6 pathologies, highlighting how disruptions in specific domains contribute to human disease.

Background: RNA-Seq is being increasingly used to identify new diagnoses and clarify the impact of variants of uncertain significance in individuals with rare disorders.

Methods: Blood based RNA-Seq has been generated for >7,800 participants with a range of rare disorders recruited to the 100,000 Genomes Project. We have identified gene expression and splicing outliers for the first 5,412 participants using the DROP pipeline. We prioritised candidate causal variants focussing on relevant haploinsufficient PanelApp genes, and outlier events overlapping with structural variants and Exomiser prioritised variants.

Results: 60% of PanelApp genes are well captured in the data (TPM>5), although this varies across individual gene panels. Significant splicing and/or expression outliers were identified in relevant PanelApp panel genes in 20% of the cohort. We will discuss how to access and interpret the RNA-Seq and outlier data in your own research.

Conclusion: Blood based RNA-Seq analysis identified candidate causal variants across a range of different rare disorders. The RNA-Seq data and identified splicing and gene expression outliers are available for NGRL researchers to aid diagnostic variant identification, explore mechanisms of rare disorders and apply multi-omics research methods.

Introduction
Genomics England’s Generation Study will sequence 100,000 newborns to screen for more than 200 rare conditions (≈500 genes). As part of the programme, an economic evaluation will assess the cost-effectiveness of genomic newborn screening (gNBS).

Objectives
To provide early, decision-oriented insight ahead of the full evaluation, we conducted a pragmatic threshold analysis to estimate the minimum health gains required for gNBS to be cost-effective at established willingness-to-pay (WTP) thresholds.

Methods
Panel wide birth prevalence for 200+ conditions was extrapolated from a sample of 50 conditions (68 genes). UK sequencing costs were identified through a literature review of Embase and MEDLINE (29 January 2026). Breakeven QALY requirements were calculated at £51,000/QALY gained, with scenario analyses at £30,000 and £100,000/QALY gained.

Results
Assuming £900–£1,600 gNBS costs per newborn and a 0.3–0.6% panel wide birth prevalence, the central scenario (£1,200; 0.44%) yields an average breakeven requirement of 5.4 QALYs gained per affected infant. Across the WTP scenarios tested, the breakeven requirement ranged from 2.7 to 9.2 QALYs per affected infant. Potential sources of health gain include earlier initiation of effective treatments and avoidance of prolonged diagnostic odysseys.

Conclusions
This analysis provides early insight into the health gains needed for gNBS to be cost-effective in England. A 5.4 QALY gain represents the ‘average’ breakeven requirement across the panel. Some conditions may yield larger gains—for example, spinal muscular atrophy, where substantial QALY benefits arise from early intervention—while others may yield smaller gains. These findings can help guide an archetype-based approach to assessing cost-effectiveness, thereby avoiding individual-condition models."

Rapid technical innovations in prenatal genomic sequencing using cell-free fetal DNA, and gene, stem cell and enzyme replacement therapies delivered in utero, could in the future dramatically alter the scope of prenatal testing and treatments for monogenic conditions. In this qualitative study, we used deliberative dialogue to ask key stakeholders – how can we screen for genetic conditions in pregnancy and deliver in-utero therapies in an equitable and ethical way in England? Data collection comprised online workshops and an optional follow-up survey with professionals (n=28) and people impacted by genetic conditions (n=35). Data were analysed using inductive content analysis. Prenatal screening and in utero therapies were largely acceptable to participants, with early diagnosis and treatment for a range of genetic conditions viewed positively. Participants wanted holistic and well-coordinated care for parents who are offered these new tests and treatments. Accurate information and support for parental decision-making about prenatal therapies was seen as a priority, as decisions would be shaped by emotional stress, tight timelines and uncertainties around condition prognosis and treatment efficacy. Gaps in current care for families impacted by genetic conditions were noted, and participants were concerned that expanding screening and therapy options would increase pressure on already stretched NHS services. Challenges to ensuring an equitable service were also highlighted, as access to screening and therapies may vary by location, patient and public understanding and attitudes and resources. Findings will inform future research, policy and guidelines to ensure acceptable translation of these new technologies into clinical practice.

The role of genomics in healthcare is expanding rapidly. Many initiatives, including the Generation Study, are exploring the benefits and challenges of using genomic newborn screening (gNBS) to expand newborn screening programmes. We aimed to explore NHS professionals’ views on gNBS to obtain an understanding of perceptions, acceptability, feasibility and workforce implications. An online survey is being distributed through professional bodies, special interest groups, conferences, key contacts and social media between 14/11/2025 and 31/03/2026. 603 surveys have been completed to date. Participants include midwives (n=98), nurses (n=47), health visitors (n=45), clinical geneticists (n=54), genetic counsellors (n=33), laboratory scientists (n=46), research midwives/nurses (n=68), paediatricians (n=43) and general practitioners (n=70). Almost all were aware of current newborn screening (98%) and most were aware of the Generation Study (72%). The mean acceptability score for offering gNBS in the NHS was 16.6 (minimum and maximum possible scores: 4 and 20). Participants scored the importance of common benefits and risks of gNBS from 0 to 10. Highest scoring benefits were: monitoring the child so they can receive timely treatment (Mean 9.1) and avoidance of diagnostic delay (Mean 9.0). Highest scoring risks were: potential long-term impacts of uncertain results on families (Mean 7.8) and false positive results (7.2). Highest ranked challenges for offering gNBS routinely were: accessible and equitable access to gNBS and appropriate staffing of clinical services. Overall, whilst most participants were supportive of gNBS, there are concerns around uncertain findings, equity of access and workforce that need to be addressed prior to routine implementation.

Polygenic risk scores (PRS) are quantitative summaries of a person's genetic predisposition to a disease or trait. While PRS have shown considerable potential for clinical utility, for many traits their predictive performance remains starkly uneven across diverse genetic ancestries. To improve transferability, numerous ancestry-aware approaches have been proposed, many of which aggregate models trained within separate genetic ancestry groups or adjust for group-specific differences. However, these approaches typically rely on discretising genetic ancestry into categorical groups, often using continental labels. Such discretisation risks conflating biological variation with sociopolitical constructs such as race and perpetuating discriminatory belief systems.
Here, we explore efforts to move towards continuous characterisations of genetic ancestry for PRS. We then examine the use of kernel methods to this end and propose a kernel-based formulation of genetic distance, which naturally captures smooth genetic variation without imposing discrete ancestry boundaries. This aims to implement contemporary conceptualisations of genetic ancestry, and may improve generalisability of PRS across heterogeneous populations.

Rare genetic diseases remain a major diagnostic challenge, with many patients undergoing prolonged diagnostic odysseys despite access to genomic testing. Improving diagnostic yield requires both effective variant prioritisation and robust strategies for translating genomic data into clinically actionable findings. Here, we use craniosynostosis - the premature fusion of sutures between the skull bones - as a model condition to develop and apply approaches for novel disease–gene discovery within genome sequencing data from the Genomics England NHS Genomic Medicine Service (GMS).

This study spans two partially overlapping cohort releases (GMS v4 and v5), comprising over 150 families with craniosynostosis, and includes independent analysis of sequence and structural variation. Focusing on molecularly unsolved cases, we use trio-based analysis, Exomiser prioritisation, and gene constraint metrics (among other methods) to identify rare, potentially pathogenic variants. Preliminary analysis of 63 trios has yielded five candidate genes, including at least one emerging disease–gene association supported by recent external evidence.

We will present representative findings that emerged from our analysis. For each case, we outline key considerations, decision points and follow-up experiments, if pertinent. We also introduce a framework for prioritising variants for downstream functional validation, including selection of the best variants for animal modelling.

Finally, we highlight key challenges in working with this data, including limitations in clinical follow-up, and consider their impact on gene discovery.

We hope that these strategies contribute to improving the diagnostic yield and advancing the resolution of the diagnostic odyssey for patients with rare genetic diseases.

Rare genetic diseases remain a major diagnostic challenge, with many patients undergoing prolonged diagnostic odysseys despite access to genomic testing. Improving diagnostic yield requires both effective variant prioritisation and robust strategies for translating genomic data into clinically actionable findings. Here, we use craniosynostosis - the premature fusion of sutures between the skull bones - as a model condition to develop and apply approaches for novel disease–gene discovery within genome sequencing data from the Genomics England NHS Genomic Medicine Service (GMS).

This study spans two partially overlapping cohort releases (GMS v4 and v5), comprising over 150 families with craniosynostosis, and includes independent analysis of sequence and structural variation. Focusing on molecularly unsolved cases, we use trio-based analysis, Exomiser prioritisation, and gene constraint metrics (among other methods) to identify rare, potentially pathogenic variants. Preliminary analysis of 63 trios has yielded five candidate genes, including at least one emerging disease–gene association supported by recent external evidence.

We will present representative findings that emerged from our analysis. For each case, we outline key considerations, decision points and follow-up experiments, if pertinent.

We also introduce a framework for prioritising variants for downstream functional validation, including selection of the best variants for animal modelling. Finally, we highlight key challenges in working with this data, including limitations in clinical follow-up, and consider their impact on gene discovery.

We hope that these strategies contribute to improving the diagnostic yield and advancing the resolution of the diagnostic odyssey for patients with rare genetic diseases.

Objective: This study develops and applies an implementation costing framework to estimate the resources and costs required to deliver genomic newborn screening (gNBS) at scale in England.

Methods: We developed a conceptual framework to estimate the costs of implementing gNBS within a population screening programme. The framework draws on theories of innovation adoption, scale, and organisational costs to examine how programme uptake, service capacity and workforce requirements influence costs. Factors affecting implementation were identified using the Consolidated Framework for Implementation Research (CFIR). Data informing the framework include a national survey of healthcare professionals designed to identify priority implementation strategies, workforce training needs, and operational requirements for integrating gNBS into routine care. The Cost of Implementation Strategies (Cost-IS) instrument captures the resources required to deliver these activities. These data will inform a future budget impact analysis estimating the cost of scaling gNBS across England. Scenario and sensitivity analyses will explore uncertainty in key parameters, including sequencing capacity, workforce training, programme coordination, and variation in uptake across sites and population groups.

Results: Preliminary mapping identified factors likely to influence implementation costs, including equitable access across populations, sufficient clinical workforce capacity, and clear clinical pathways supported by training in genomic result interpretation. These factors will inform the specification and costing of implementation activities.

Conclusions: This framework provides a structured approach to identifying and costing activities required to scale gNBS. Our findings will inform budget impact analyses and support strategic planning, resource allocation, and equitable integration of genomic screening into routine newborn care.

Objective: This study assesses out-of-pocket costs and parent-reported outcomes following genomic newborn screening (gNBS).

Methods: As part of the independent Process and Impact Evaluation of the Generation Study in England, a survey was administered to parents approximately three months after receipt of their child’s gNBS results. The survey collected information on healthcare utilisation and family out-of-pocket costs for follow-up appointments, over-the-counter medications, medical equipment, and dietary or therapeutic products. It also captured productivity impacts (presenteeism and absenteeism), childcare and schooling expenses and informal care requirements. Health-related quality-of-life was measured using validated instruments: the EQ-5D-5L for parents, EQ-TIPS for infants, and the GAD-7 for parental anxiety. Non-health outcomes included perceived utility of genomic information, decision regret, the Vulnerable Baby Scale, and the Mother-to-Infant Bonding Scale. Descriptive analyses were conducted to summarise reported costs and parent and infant outcomes among respondents.

Results: Data collection is ongoing, with 251 parents completing surveys (233 no suspected conditions (NCS), 18 with condition-related outcomes). Follow-up for newborns NCS will conclude in April 2026. Data collection for the cohort with confirmed conditions will continue until completion of the Generation Study. Preliminary analyses will report descriptive statistics on healthcare utilisation, out-of-pocket costs, productivity impacts, and parent- and infant-reported outcomes, including comparisons between families with and without suspected or confirmed conditions, where sample sizes permit.

Conclusions: Understanding family-level costs and outcomes is essential for comprehensive cost-effectiveness evaluations of gNBS. These findings will inform future economic modelling and support policy decisions on potential implementation of gNBS within national screening programmes."

Background: Despite advances in genomic sequencing, diagnostic rate of rare diseases remains less than 25%. Systematic approaches integrating gene prioritisation, phenotype correlation, and variant interpretation are needed to identify new genetic diagnoses.

Methods: We applied a gene-led approach using the GenePy score, which aggregates variant functional impact into a per-gene pathogenicity burden for each individual. Genomic data from ~78k participants from ~34k families affected by rare diseases in the Genomics England, were transformed into a GenePy matrix. A multi-tiered workflow was developed to link GenePy score to phenotype association leading to diagnostic evaluation. For each gene in the genome, probands with top-ten GenePy scores were assessed for gene–phenotype based on the diagnosis and manifested HPO terms. Positive association genes were then reviewed on the mutation profile in the family for variant clinical pathogenicity, segregation, and inheritance mode. The process was automated and iteratively refined into a clinic-friendly pipeline, with cases reviewed by a multidisciplinary team.

Results: Gene–phenotype correlations were identified in 5,272 probands, where the proband carried a top-10 GenePy score of a gene and manifested the related phenotypes of the signpost gene. This included 2,548 family who have yet received a genetic diagnosis, of whom 421 cases carried pathogenic variants meeting ACGS-2024 diagnostic criteria and is under clinical evaluation following multidisciplinary review.

Conclusions: This gene-first framework, combining computational prioritisation with phenotype-driven validation, improves diagnostic yield and supports discovery of novel disease genes. The approach is scalable and applicable to other large rare disease cohorts."

Anophthalmia and microphthalmia (A/M) are rare developmental eye disorders characterised by the absence or reduced size of the eye. These conditions are genetically heterogeneous, with over 140 causative genes identified, yet much of their genetic basis remains unresolved, resulting in only 20-30% of affected individuals receiving a genetic diagnosis. This highlights a critical gap in understanding the genetic basis of A/M and the need to identify additional causal variants to improve diagnosis and understanding of disease.

We investigated unresolved A/M in participants recruited to the 100,000 Genomes Project to identify novel causal genes and improve diagnostic rates. Analysis focused on families with available trio or extended family data, particularly those with more than one affected individual, enhancing segregation analysis. A broad range of variant types was assessed and prioritised based on low allele frequency and predicted functional impact.

Eight candidate variants were identified in two probands. Notably, a pathogenic variant in KIF17 showed strong segregation with disease and a high predicted impact on splicing. As KIF17 has only been associated with microphthalmia in one reported family, this study strengthens the evidence supporting its role in A/M.

Our ongoing work aims to functionally validate candidate genes by CRISPR knockdown in zebrafish to assess their role in eye development and investigate the impact of the identified variants in vivo, providing further insight into disease mechanisms. Ultimately, this study has the potential to increase molecular diagnosis rates among affected individuals and improve our understanding of A/M.

Understanding the genetic architecture of complex disease requires models that capture non-linear relationships between variants, not just additive effects. Large-scale whole genome sequencing (WGS) datasets, such as the UK Biobank, enable this at population scale, provided appropriate computational methods and infrastructure are available. Trusted Research Environments (TREs) offer secure, scalable access to such data, but their implications for advanced analytical methods remain underexplored.

We deployed VariantSpark and BitEpi, our Spark-based machine learning (ML) GWAS framework, on UK Biobank WGS data within the Research Analysis Platform (RAP) to identify non-linear genetic associations and epistatic interactions for two phenotypes, atrial fibrillation (n=37,000) and type 2 diabetes (n=40,000) across 8 million variants.

Deploying complex ML workflows, such as VariantSpark, in the RAP is not plug-and-play. Performance did not scale linearly with compute, with diminishing runtime gains and substantial cost. Runtime and cost were strongly influenced by data format (e.g. PLINK vs VCF) and repeated data access from storage. Through iterative optimisation of the VariantSpark workflow, we reduced runtime by 52% to 10 days and cost by 27% to ~£1,800 per phenotype, noting that infrastructure constraints somewhat limit optimisation.

Having established a scalable, cost and runtime-optimised framework, we have completed end-to-end screening across both phenotypes. Full association results, candidate loci, and epistatic interactions will be presented at the summit. Outputs are being integrated into an Ensembl VEP plugin for broader, open dissemination. This work highlights that realising the promise of WGS for unravelling complex disease requires co-evolving analytical methods with TRE infrastructure.

Human genetics has become a cornerstone of drug target discovery, yet the value of Mendelian randomization (MR) for predicting clinical success remains uncertain. Here, we systematically evaluated MR across 11,482 target–indication pairs with documented Phase II clinical outcomes to assess its utility for drug development. We find that MR statistical significance alone does not enrich for Phase II success, in contrast to genome-wide association study (GWAS) support, which confers an increase in success probability. However, this apparent limitation reflects the heterogeneous nature of clinical failure and the fact that MR encodes information beyond P values. When MR-derived features, including instrument strength (F-statistic) and explained variance (R2), are integrated into machine learning models, predictive performance improves substantially. An MR-informed XGBoost classifier identifies target–indication pairs with a 55% overall approval rate, corresponding to a 6.4-fold enrichment over unstratified programs and a 2.8-fold improvement over GWAS-supported targets in Phase II. Notably, this enrichment is achieved without reliance on statistically significant MR results. Our findings demonstrate that MR is most informative when treated as a graded, context-dependent source of causal evidence rather than a binary hypothesis test, and that its integration with machine learning enables scalable, genetics-informed prioritization of drug targets across the clinical pipeline.

Background: Standard HER2 testing (IHC and ISH) captures protein expression and gene amplification but not downstream signalling. Since H&E morphology reflects underlying tumour biology, AI models trained on these images may implicitly encode functional HER2 activity. We hypothesised that clinically HER2-positive tumours classified as negative by AI are enriched for inactive HER2 signalling, suggesting standard diagnostics overestimate true pathway activity.

Methods: A deep learning AI model was applied to 888 TCGA breast cancer cases with H&E images and clinical HER2 results. Cases were stratified by concordance as follows: AI+/HER2+ (n=91), AI−/HER2+ (n=82), AI+/HER2− (n=175), and AI−/HER2− (n=540). Cases were characterised by ERBB2 z-scores (RPPA/RNA-seq) and a HER2 signalling class (Active/Partially Active/Inactive) based on ERBB2 and 11 downstream effectors across PI3K/AKT, mTOR, and MAPK axes.

Results: AI−/HER2+ cases occupied a distinct intermediate molecular stratum: ERBB2 expression was significantly lower than in AI+/HER2+ cases (median: 0.38 vs 1.51; p=0.003) yet higher than AI−/HER2− cases (median: −0.25; p<0.001). Sixty percent of AI−/HER2+ cases were HER2-Inactive compared to 40% of AI+/HER2+ cases (p=0.01), indicating AI morphology aligns better with pathway status than clinical labelling alone.

Conclusions: AI identifies a clinically HER2-positive subgroup with inactive signalling, suggesting standard testing may overcall molecular activity. Flagging these patients offers a biological rationale for molecular reassessment and consideration of HER2-low therapies such as trastuzumab deruxtecan prior to treatment decisions.

Immune related adverse events (irAEs) following treatment with immune checkpoint inhibitor (ICI) therapy remain a critical challenge, with myocarditis being especially rare, unpredictable and often fatal. Predicting risk of ICI induced myocarditis therefore remains a crucial approach to improving outcomes. Thus far this has been hindered by a lack of predictive biomarkers, incomplete mechanistic understanding and under-reporting/diagnosis. Recently, the emergence of pre-existing autoreactive T cell clones as a key initiator suggests a mechanistic link to autoimmune disease. Given the strong genetic predisposition in autoimmunity, we hypothesize that rare genetic variants predisposes certain patients to this severe irAE. To aaddress this gap, we conducted a retrospective case-control analysis leveraging data from the National Genomic Research Library and the Identification of genetic factors that predispose to immune checkpoint inhibitor toxicity (ICI Genetics) research study. Cases were defined as patients who developed myocarditis following checkpoint inhibitor therapy, with controls referring to patients exposed to ICIs but who did not develop irAEs. A preliminary scoping review of current literature on myocarditis was conducted to identify and prioritise candidate genes for analysis. We performed a rare variant burden test to filter for rare (MAF < 0.01), non-synonymous and loss of function variants within these prioritised genes to evaluate for protective or predisposing polymorphisms between cases and controls. By identifying genetic predictors of myocarditis, this pilot study aims to improve risk stratification to enable safer immune checkpoint inhibitor application.

Transcript choice is a critical but underappreciated determinant of clinical variant interpretation. Historical discrepancies between RefSeq and Ensembl/GENCODE annotations have complicated harmonisation of reporting, prompting the MANE project to define a single MANE Select (MS) transcript per gene, alongside MANE Plus Clinical (MPC) transcripts capturing clinically relevant alternative isoforms. However, the extent to which MPC transcripts influence diagnostic interpretation remains incompletely characterised.

We analysed 65 genes in MANE v1.4 with both MS and MPC transcripts, comparing transcript structures and defining transcript-specific coding sequence using UCSC Genome Browser multi-region visualisation and interval-based analyses. We developed an interactive browser resource (1,863 regions; 717 kb) enabling simultaneous multi-transcript inspection (https://github.com/ExeterGenetics/MANE_multiview/tree/main). Read coverage (gnomAD v4) and pathogenic variants from ClinVar and the NGRL were integrated to assess clinical impact.

Structural differences included alternative first exons, mutually exclusive exons and complex rearrangements. The median unique coding sequence per MS/MPC pair was 115bp, with outliers such as DST and SYNE1 (>20kb). Approximately 2.5% of (likely) pathogenic variants mapped exclusively to MPC transcripts, indicating potential under-recognition if only MS transcripts are considered. MPC-specific exons in BRAF, SYNE1 and REEP6 showed poor exome coverage, highlighting diagnostic blind spots. In some genes (e.g. IQSEC2, MOCS2), MS and MPC transcripts encode entirely distinct proteins.

Overlapping, opposing-strand start codons (e.g. MUTYH/TOE1) suggest under-recognised mechanisms for phenotype blending.

These findings demonstrate that systematic consideration of MPC transcripts is important for accurate rare disease variant prioritisation. Our visualisation framework also provides an accessible and scalable resource for training and education in clinical genomics.

Reporting of secondary findings (SF) from WGS remains controversial. The American College of Medical Genetics and Genomics recommends active interrogation of actionable genes with patient opt-out, while European guidance advocates greater caution, particularly in children and for late-onset conditions. Within the NHS GMS, SF were evaluated in the 100,000 Genomes Project but are not routinely reported in current WGS pathways. However, phenotype-driven variant prioritisation tools, such as exomiser, may nonetheless highlight pathogenic variants unrelated to the primary indication.
We evaluated the frequency of SF identified in patients diagnosed with the rare autoimmune bleeding disorder, immune thrombocytopenia (ITP). WGS was performed in 130 patients (100 adults; 30 children) with ITP to investigate inborn errors of immunity (R15) or hereditary thrombocytopenia (R90), using standard NHS GMS pipelines.
Pathogenic SF unrelated to ITP were identified in 1% of adults (MSH6) and 10% of children (TP53, PALB2, RYR1), predominantly involving cancer predisposition genes. Additionally, variants explaining the presenting phenotype but conferring malignancy risk were found in 4% of adults (NFKB1, NFKB2, ANKRD26) and 3% of children (ETV6), blurring the boundary between primary and secondary results. SF were reviewed by a Clinical Genetics MDT with case-by-case disclosure decisions, reflecting the absence of a national framework. These findings highlight the need for robust consent processes, consistent disclosure policy, structured MDT involvement including clinical geneticists and genetic counsellors, and adequate resourcing for post-test counselling. As genomic testing becomes embedded in the NHS, the uncertain implications of SF for patients, families, and health services cannot be overlooked.

Background: As ML models are increasingly used in genetics, a key question is whether these approaches outperform traditional PRS models; alongside the critical role of appropriate control and covariate selection for robust genetic-AUC estimation. To answer this, we set out a systematic comparison of PRS, Artificial Neural Network (ANN), explainable ANN, and random forest methods.

Methods: The study included 3,956 individuals with AD and three different control sets of equal size: 1) randomly selected; 2) matched on technical covariates and 3) full matched further including age and sex. UK Biobank genotypes (~96M SNPs) underwent strict quality control and Linkage Disequilibrium (LD) pruning yielding ~39K SNPs. SNPs associated with AD in an independent dataset (Lambert et al, 2013) were used for feature selection at multiple P-value thresholds. Explainable ANN models (60:20:20) were trained using GenNet, with gene level annotations derived from Ensembl resources. PRS was performed by PRSice-2.

Results: Prediction accuracy was highest when full covariates were included, with PRSice-2 achieving AUC = 0.825 in the unmatched set and AUC = 0.815 in the technically matched set. In ANN, accuracy improved when pre selected SNP features were used, achieving AUC = 0.795, lower than PRS (AUC=0.806). Interpretability analyses of ANN highlighted established AD associated 19q13.32 locus. Overall, in this preliminary work, PRS outperformed the biologically informed ANN framework.

Discussion: This study provides one of the first direct comparisons between GWAS based PRS and ANN models, and future work will extend this methodological framework and its applications in Genomics England dataset.

Funding: Black Scholars Phd studentship.

BACKGROUND Post-translational modifications (PTMs) regulate protein function and disease, and can now be predicted from coding sequences using machine learning, enabling translation of genetic variation into clinically relevant insights for personalised medicine.

METHODS We predicted PTMs in exonic sequence of GLP1R gene harboring rs10305492 (A316T) known to effect random glucose (RG) levels, in 450K UKBiobank individuals. Pancreatic tissue PTM sites of GLP1R were obtained from curated databases (dbPTM/qPTM/PhosphoSitePlus/UniProt) and modelled using sequence-based learning approaches (Random forest, X-gradient boosting, linear regression, support vector machines) to predict site-specific probabilities in training set of experimentally validated PTM sites and matched negative samples. We used the best performing parameters by X-gradient boosting to predict the probability of PTMs within 678 unique GLP1R proteoforms we identified. PTM probability features were then mapped to individual-level data to capture potential functional effects on RG by linear regression models.

RESULTS This effort yielded probabilities for nineteen previously known PTM sites and seventeen previously undetected PTM sites to be investigated further. Notably, we identified eight PTM sites associated with A316T including the established phosphorylation gain/loss at PTM(316T) (P<1.10E-300), however none of these PTMs associated with RG significantly.

DISCUSSION This represents a first attempt to evaluate the utility of in silico–predicted PTMs in a large population setting. The approach should be validated in external datasets such as Genomics England. One limitation is that existing PTM databases do not include brain tissue for GLP1R, which may have provided more biologically relevant insights.

Climate change is one of the greatest contemporary challenges to human health, undermining human health through multiple mechanisms. Among relatively understudied mechanisms are those related to individual genomic variation. We aimed to examine this possibility.

Through a defined, agnostic literature review-based approach, we curated human genetic variants with functionally characterised temperature-dependent effects: we call these ‘calortypic variants’, some of which are linked to temperature-sensitive disease phenotypes. Next, we examined their occurrence in whole-genome sequenced rare disease cohort and analysed their associated phenotypes. Finally, we performed transcriptomic analysis in astrocyte models to examine the impact of short-term exposure to elevated ambient temperature.

A set of 159 calortypic variants across 65 calortypic genes was identified; most (66.7%) calortypic variants caused temperature-sensitive disease phenotypes, and 44.7% were found in neurological and neurodevelopmental diseases. Calortypic variants were also found in 300/39 834 participants recruited to the Genomics England (GEL) 100 000 Genomes rare disease programme.

Temperature-related phenotypes were documented in eight GEL participants; in 6/8 participants (two probands and four of their relatives), calortypic variants had already been identified as the disease-causing variant. Gene expression changes across human astrocyte transcriptomes under different temperature exposures prominently involved genes associated with biological processes implicated in a range of neurological diseases.

Genetic variation may generate latent phenotypes that manifest only at elevated ambient temperatures, with some neurological disease groups being highlighted. This is an exploratory study. Identifying more calortypic variants will help uncover the full spectrum of human genetic vulnerability to climate change impacts.

Background: Common variants are associated with risk of rare, neurodevelopmental conditions (NDCs), but do not only exert direct genetic effects. NDC trio analyses show effects of non-transmitted parental alleles, possibly reflecting genetic nurture or assortative mating. Since complete trios are biased toward families with higher socioeconomic status, ascertainment bias could confound these results. Mendelian imputation, which predicts genotypes for the missing parent in duos, could reduce bias and improve power. Here, we harness this method to assess direct and non-transmitted effects of common variants on NDC risk.

Methods: Combining imputed (n=8768) and observed trios (n=7282) from two NDC and two birth cohorts, we calculated polygenic indices (PGIs) for NDCs, education attainment (EA), its cognitive and non-cognitive components, and schizophrenia. We estimated direct and non-transmitted effects using the trio model.

Results: Unlike previous work showing no direct PGI-EA effect on NDC risk, including imputed trios resulted in a nominally-significant positive direct effect of EA (p-adj>0.05; p<0.05) and its non-cognitive component (p-adj>0.05; p<0.05). This effect was driven by imputed trios in birth cohorts, which have significantly lower EA PGI than observed trios. It is unclear how this ascertainment could create a positive direct effect, since we expect it to influence only non-transmitted effects. The effects of PGIs for other NDC-related traits were largely unchanged.

Conclusion: Including imputed trios unexpectedly revealed a positive direct effect of PGI-EA on NDC risk. We are exploring potential explanations, including ascertainment bias on the child independently of the parents or common-rare variant correlations due to assortment.

MED13L syndrome is a rare autosomal dominant disorder characterised by developmental delay/intellectual disability (DD/ID) and other variable features. This study aimed to refine phenotypic variability based on the genotype.
In the first phase, 183 individuals with MED13L syndrome were studied including 119 previously published and 64 new cases collected through international collaborations. Additionally, two cases were identified through the 100,000 Genomes Project.

Across the 183 individuals (12 months–63 years old), 139 harboured likely gene-disrupting (LGD)/splice-site, 42 missense, and two whole-gene gain variants. The largest concentration of variants was on MED13L C-terminus. Notably, motor delay was absent in 3% of cases (one missense, four LGD/splice-site), but speech delay was present in all (14% mild, 38% moderate, 48% severe-profound) except one individual with a whole-gene gain. Compared with LGD/splice-site, individuals with missense variants more frequently presented with seizures (61% vs 17%, p=1.01×10⁻⁵), as well as severe-to-profound motor delay (38% vs 14%, p=0.007) and speech delay (68% vs 43%, p=0.03). Missense variants showed concentration and recurrence around a degron area (aa844–869) and C-terminus (aa1899-2195). Individuals harbouring degron missense variants tended to exhibit severe/refractory seizures, more severe DD/ID, and one stillborn. We identified two likely/pathogenic MED13L missense variants from the 100,000 Genomes Project, further expanding the genotype-phenotype correlation.

We propose three potential subgroups within MED13L syndrome: (A) haploinsufficiency (LGD/LoF missense), (B) possibly dominant-negative effect (certain missense variants especially between aa844–869), and (C) whole-gene gain with mild presentation (likely underreported). These findings have implications for variant classification, clinical prognosis, and mechanistic investigation.

The 100,000 Genomes Project (100KGP) undertook genome sequencing of patients with rare diseases and cancer to study the role that genes play in disease, and to integrate genomics into UK healthcare. To contribute to 100KGP, the Wales Genomic Medicine Centre (a partnership between the NHS All Wales Medical Genomics Service (AWMGS), Cardiff University and Genomics England) recruited 438 individuals from 154 families that had been subject to pre-genomic genetic testing without reaching a diagnosis. The majority of probands (64%, 98/154) had neurological or neurodevelopmental phenotypes. Genome sequencing, variant calling, gene-based filtering and variant prioritisation were performed by Genomics England. AWMGS undertook clinical interpretation, validation and reporting of variants. Initial diagnostic yield was 20.8% (32/154) with variants of uncertain significance (VUS) reported for 11 more families. Most of the initial findings (81.4%, 35/43) could have been detected by clinical exome sequencing which was standard of care in Wales at the time. Reanalysis of the 100KGP Wales data using updated variant prioritisation tools, expanded gene lists, re-phenotyping and segregation studies, has increased diagnostic yield to 42.2% (65/154) (an improvement of 103%). RNA analysis was used to clarify the clinical significance of VUS in COQ4, ENPP, GATAD2B, NCAPD2, and THOC2. These findings demonstrate the clinical utility of genome sequencing, RNA analysis, and periodic reanalysis of genomic data.

Neuroblastoma (NB) is the most common solid paediatric tumour outside the central nervous system. Clinical outcomes are highly heterogeneous: low-risk disease may undergo spontaneous regression, whereas high-risk disease, accounting for approximately 50% of cases, is associated with substantial morbidity and mortality. Standard of care (SOC) diagnostics comprises MYCN FISH, SNP arrays and short-read whole genome sequencing. These provide key insights into genomic abnormalities for risk stratification but have significant limitations when resolving complex structural variants in NB, for example within telomeric and sub-telomeric regions.

Telomere maintenance mechanisms (TMM), including TERT alterations and alternative lengthening of telomeres, are being investigated as possible risk defining structural variants in NB, but current SOC diagnostics struggle to detect TMM. Oxford Nanopore Technologies (ONT) long-read WGS presents a single methodology to characterise both risk defining and novel structural variants including telomere length.

We will evaluate sequencing data of six tumour:normal NB sample pairs produced by Genomics England. Median genome coverage of these samples is 39.7X (±14.02) for germline and 54.3X (±11.8) for tumour samples. Corresponding median N50s are 11.7kb (±4.7kb) and 10.8kb (±4.7kb), respectively. In parallel, we will complete ultra-long ONT sequencing on three NB samples of differing TMM status, targeting an N50 of >70 kb.

Through analysis of the ultra-long ONT data, we aim to assess telomere length, correlating this information with TMM status. Additionally, using this analysis alongside the Genomics England data, we will demonstrate the clinical utility of ONT sequencing in NB diagnostics, benchmarking its detection of genetic abnormalities against SOC testing.

Trusted Research Environments (TREs) are increasingly central to national genomic initiatives such as Genomics England, providing secure, auditable and privacy-preserving access to clinical and genomic datasets. Extracting biological insight from these population-scale resources requires analytical-tools that operate fully within TRE governance while delivering advanced causal-inference and multi-omics capabilities.

bioXcelerate has developed DiscoveryX, a modular causal genomics toolkit designed for TRE deployment, including UKB-RAP. DiscoveryX integrates scalable high-resolution finemapping (SwitchStep), multi-trait colocalisation (Pleiograph), summary statistic imputation (ImpMap) and Mendelian Randomisation. These tools have demonstrated substantial performance gains, reducing genetic variant detection times by up to 98% and accelerating finemapping and colocalisation workflows from days to hours or minutes, enabling secure and reproducible analysis at scale.

We applied DiscoveryX to a large Parkinson’s disease (PD) use case using summary statistics from a major European GWAS meta-analysis (Nalls et al., 2019) and integrated more than 4,200 publicly available molecular QTL datasets spanning eQTLs, pQTLs, sQTLs and metabolite-QTLs across blood and brain. DiscoveryX identified over 30 molecular traits with causal links to PD risk, including genes, proteins, metabolites and splice variants. Embedding these signals into a PD-specific knowledge graph revealed biological processes newly implicated in PD pathogenesis.

Protein‑coding candidates such as EFNA1 and HIP1R emerged as causally linked to both PD and cardiometabolic or lipid‑related traits, suggesting shared mechanisms across neurological and systemic biology. Incorporating a clinician‑curated set of candidate therapeutic targets further revealed mechanistic convergence on pathways such as calcium signalling and fibroblast growth factor (FGF) biology, pointing to additional routes for validation.

Endometrial cancer (EC) is the most common gynaecological malignancy in high income countries, and is increasing in incidence. Molecular stratification has improved EC management; however, precision care is hampered by incomplete characterization of the EC genome. We addressed this by analysis of whole genome sequencing (WGS) of 665 ECs performed by the UK Genomics England 100,000 Genome Project (100KGP). 5% of cases were associated with germline pathogenic variants in cancer genes, including BRCA1 which we confirmed predisposes to EC by case-control study. We identified 107 putative coding driver genes, 35% of which had no prior established role in EC. Novel structural variants included gains of MYCN and loss of its negative regulator NEDD4.1 which were significantly mutual exclusive in copy number (CN) high tumours. Immunogenomic analysis confirmed selection for driver alterations of low immunogenicity based on patient HLA haplotype, and pervasive immune escape through multiple mechanisms. Unsupervised clustering of mutational signatures and genomic alterations identified known and novel molecular subgroups, including a CN-high subset with mutational signatures of homologous recombination deficiency (HRD) and favourable outcome. Independent prognostic value of single nucleotide variant (SNV) burden, CN burden and multiple coding drivers, along with the identification of targetable molecular alterations in over one-third of cases, underscored the promise of WGS for precision medicine.

Population scale analyses are essential for characterising clinically relevant recurrent genomic alterations in cancer, yet existing somatic variant frequency resources are often heterogeneous and lack standardised assay scope and sample types. To address this, we generated whole genome sequencing (WGS)-based somatic variant frequency aggregates for solid and haematological malignancies using data from the 100,000 Genomes Project and the NHS Genomic Medicine Service (GMS).

Somatic small variants were derived from paired tumour-germline WGS data and aggregated using a consistent analytical pipeline. The current release includes non synonymous and splice region variants across more than 600 cancer genes, summarised at both coding DNA sequence (CDS) and protein levels. The dataset comprises over 11,000 solid tumours spanning 31 tumour types and approximately 2,300 haematological samples from major lymphoid and myeloid disease groups, and includes over 1,100 variants observed in more than ten cases. All variants were annotated against MANE transcripts and supplemented with quality flags.

To support application of SVIG UK guidelines, we computed additional variant level summary metrics, including counts of downstream truncating mutations and nested in frame deletions. Whole genome aggregation enables interpretation of variant frequencies alongside absolute variant counts through a uniform test scope and provides a framework for analysing non coding variants and additional variant classes.

As WGS uptake in cancer remains limited, we are exploring integration of targeted panel data while accounting for differences in test coverage to increase resolution without compromising frequency interpretability. These resources provide a scalable, standardised framework for population scale assessment of somatic variant frequency in cancer genes.

Understanding how recruitment patterns and genetic factors influence access to genomic diagnosis is essential for ensuring equity. However, the relationship between socioeconomic deprivation and diagnostic yield in rare developmental disorders (DD) remains unclear. Using integrated genetic, clinical, and socioeconomic data from the 100,000 Genomes Project, we analysed diagnostic outcomes in 16,138 DD participants (8,823 trios; 7,315 non-trios) and assessed variation in inheritance patterns across deprivation deciles defined by the Index of Multiple Deprivation (IMD).

Participants from the most deprived areas were over-represented (12.76% in IMD decile 1 vs 10% expected; p=1.54×10⁻²⁹) and less likely to be recruited as trios. Overall diagnostic yield was 20% and did not differ significantly across IMD deciles. However, yield was higher in trios than non-trios (22% vs 17%; p=5.04×10⁻¹¹). While overall yield was consistent, the mode of inheritance varied: de novo diagnoses were more common in less deprived groups, whereas homozygous diagnoses were enriched in more deprived groups, largely reflecting differences in ancestry and autozygosity.

After adjusting for de novo mutation burden, autozygosity (FROH), and ancestry, increasing deprivation remained associated with a reduced likelihood of de novo diagnosis (OR 0.965 per decile, p=0.002), equating to ~14% fewer diagnoses than expected. Homozygous diagnoses were strongly associated with autozygosity (OR 1.18 per 1% FROH, p=1.8×10⁻⁴⁹), with South Asian ancestry independently contributing increased risk (OR 2.05, p=4.94×10⁻⁵).
These findings highlight how both genetic and non-genetic factors shape diagnostic patterns despite similar overall yield.

Background:
Ηuman endogenous retroviruses (HERVs) occupy ~8% of the human genome. Within the HERV-K (HML-2) group, most integrations are solo long terminal repeats (solo-LTRs), classified into three phylogenetic subgroups: LTR5Hs, LTR5A and LTR5B. Their sequence variability is poorly characterized. Here, we investigated SNP rate variation across HML-2 solo-LTRs in the human population.

Material and Methods:
Small variants (<50 bp) within solo-LTRs were identified using an aggregated dataset from the 100,000 Genomes Project, including germline data from 78,195 individuals. For each locus, SNP rate was calculated as variants per base pair; for multiallelic variants, position-specific variants were used for this calculation. Correlations between SNP rate and solo-LTR features were assessed using linear regression. Within each of the four ancestry groups (European, East Asian, South Asian, African), ancestry-unique SNP rates were calculated and compared across solo-LTR subgroups. No direct between-population comparisons were performed.

Results:
Solo-LTRs showed heterogeneous SNP rates (median 0.202; range 0.045-0.998 variants/bp). LTR5A loci had higher SNP rates than LTR5Hs (β=0.060, p=0.0004) and the most ancestry-unique variants, consistently across all super-populations. SNP rate was negatively correlated with solo-LTR age (β=−0.007, p=1.65x10-11). Intragenic solo-LTRs overlapping exonic untranslated regions (UTRs) showed higher SNP rates than non-exonic loci (β =0.056, p=0.0325).

Conclusion:
This population genomics analysis reveals genetic variability across HML-2 solo-LTRs, affected by solo-LTR age, subgroup and genomic context. Increased SNP rates observed in UTR-overlapping loci suggest a potential regulatory role for solo-LTRs embedded adjacent or inside genes. These results provide a population-level basis for prioritizing HERV-K loci for future functional studies."

mTOR pathway diseases comprise a group of 14 rare genetic disorders caused by dysregulation of the mTOR signalling pathway, which regulates cellular growth, metabolism, and survival. These conditions affect approximately 10,000 individuals in the UK and present with highly heterogeneous clinical features, ranging from multi-organ tumours to neurodevelopmental disorders such as epilepsy. Despite a shared molecular basis, patients are often managed across multiple clinical specialties, and phenotypic classification within large genomic datasets is limited by non-specific coding systems.

This study leverages linked genomic and electronic health record (EHR) data from Genomics England to refine phenotypic definitions and investigate genetic architecture across mTOR pathway diseases. We will evaluate phenotypes associated with ICD-10-coded participants by integrating multiple data sources, including Human Phenotype Ontology (HPO) terms and secondary care data, to derive a curated cohort of probable and confirmed cases. Associated genes and variants will be assessed against established mTOR pathway genes, with variant pathogenicity evaluated using ClinVar, VEP annotations, AlphaMissense scores, and GMSA tiering, and compared with unaffected relatives.

To explore phenotypic variability, we will investigate gene–gene interactions and potential genetic modifiers within the mTOR pathway. A gene-first approach will additionally be used to identify phenotypic patterns among carriers of pathogenic variants, independent of mTOR disease status. This work aims to improve phenotypic resolution in rare disease datasets and enhance understanding of the genetic and biological mechanisms underpinning mTOR pathway disorders. Findings may support improved diagnosis, variant interpretation, and future precision medicine approaches in rare disease genomics.

Background: Next-Generation Sequencing (NGS) of formalin-fixed and paraffin-embedded (FFPE) samples allows for detailed characterization of cancer from routinely collected clinical specimens. However, the preservation process often leads to significant damage and degradation of genetic material, complicating library preparation. ​​Furthermore, because sample availability is typically limited, there is a high demand for robust workflows that can efficiently convert more molecules into sequenceable libraries. Given the high variability in damage and yield across FFPE samples, generating high-quality NGS data for variant and indel identification requires a unified, high-efficiency workflow.

Methods: Presented here is a workflow leveraging the Twist TrueAmp Library Preparation and Target Enrichment workflow to maximize the conversion of damaged and low mass FFPE samples for NGS.

Results: We demonstrate library conversion of severely damaged (DIN<2) FFPE with as little as 5ng input mass. Compared to other workflows, an increase in library complexity is observed. Importantly, the enzymatic fragmentation of input DNA can be tuned to provide larger insert sizes that are appropriate for sequencing with longer read length. Coupled with target enrichment, more target sites are captured and the uniformity is significantly improved. These improvements are driven by the incorporation of Twist TrueAmp Polymerase, an engineered polymerase with a low error rate capable of amplifying through GC extremes to recover challenging genomic regions.

Conclusion: In summary, our optimized library preparation kit built on engineered enzymes emerges as a valuable asset for the deployment of NGS-based FFPE assays.

BACKGROUND:
In 2025, the 5HT2A receptor antagonist, and antiplatelet agent Sarpogrelate demonstrated equivalent efficacy in ischaemic stroke prevention compared to aspirin, without haemorrhagic side-effects (PMID:39875463). The drug is not available in the US, Europe or UK but we had previously proposed after our 2014 study (PMID:24586400) identified iron deficiency which augments platelet aggregation to 5-HT, as an ischaemic stroke risk factor in patients with pulmonary arteriovenous malformations (PAVMs). Most have hereditary haemorrhagic telangiectasia (HHT).

METHODS:
To accelerate UK introduction of 5HT2A receptor antagonists, we identified genes for pathways (biosynthesis/metabolism; kynurenine) and receptors relevant to 5-HT levels and activity, identified gene variants in controls and patients with PAVMs with and without ischaemic stroke/cerebral infarction recruited to the 100,000 Genomes Project, and evaluated gene expression in control and patient-derived blood cells.

RESULTS:
30 genes relevant to 5-HT plasma levels were identified and shown in gnomAD v2.1.1, to be subject to differential selective pressures, with reduced number of predicted loss-of-function (pLOF) variants most marked for receptor and kynurenine pathway genes. In the 24 patients with PAVMs and cerebral MRI evidence to support or refute ischaemic strokes, 91,126 variants including 65 rare pLOF variants were identified in the 30 genes, corresponding to a mean of 2.71 pLOF variants per patient. RNA-sequencing of freshly-derived platelets support comparable 5HT2A receptor and platelet gene expression in PAVM/HHT patients and controls.

CONCLUSION:
Variants in 5-HT genes could help personalise and improve real-world ischaemic stroke prevention strategies if whole genome/exome sequencing replaced current panel gene testing.

Background
The 100,000 Genomes Project (100KGP) recruited patients with specified cancers or rare diseases. Clinically-linked data enable testing of further associations with common disease. We prioritised chronic insomnia which affects 8-13% of the UK population and costs £30 billion annually due to losses in workplace productivity. Insomnia has an estimated genetic heritability of 48%, yet identifying causal genes has remained challenging.

Methods
An insomnia cohort of 385 participants was identified within the 90,178 participants in 100KGP through 13 separate SNO-MED, HPO and ICD10 codes. 978 genes identified by a 2019 genome-wide association study (GWAS) in 1.3 million people were filtered by hippocampal gene expression. Aggregated variants were compared between insomnia-enriched and non-enriched cohorts. Candidate gene expression was examined in model systems of primary human endothelial cells and peripheral blood mononuclear cells (PBMCs) treated with iron, hypoxia, bacterial or cycloheximide stress.

Results
53 candidate genes were prioritised, defining 12 Gene Ontology groups. Variants were more commonly identified in the insomnia cohort (29.1%, 112/385) than the insomnia-unenriched (20.3%, 18,214/89,315, Fisher exact p<0.0001). Frameshift and nonsense variants were identified in 6/53 genes (CRHR1, DGK1, FRS3, PTPRD, RFTN2, GRM5) where they were 6.36-fold enriched in the insomnia-enriched group (Fisher exact p<0.0005). 6 further genes demonstrated differential expression after 1h iron or 1h cycloheximide (AGAP1, DNAJC1, HS6ST3, PACRG, RANGAP1, SYT14).

Conclusion
The 100KGP dataset is a valuable resource for common, complex disease. Candidate genes reinforce the hypothesis that insomnia arises from a combination of genetic susceptibility and environmental stressors, offering targets for future therapeutic interventions."

Introduction: Patients with pulmonary arteriovenous malformations (PAVMs) are surprised to hear they increase brain abscess and migraine risks, asking what they can do to prevent these complications. We previously identified dental bacteraemias and high iron as risk factors for brain abscess, dietary factors for migraines. Our goals were to identify risk-modifying genes to potentially personalise advice and better understand mechanisms and preventative options.

Methods: PAVMs and the often associated hereditary haemorrhagic telangiectasia (HHT) were nominated as recruiting diseases for the 100,000 Genomes Project, specifying ‘brain abscess’ and ‘migraine’ in Data Models. Genome-wide association (migraine), and aggregate variant analysis using SAIGE-GENE+ (both phenotypes) were performed, focussing on endothelial function, blood-brain barrier integrity, and immune function gene variants. Primary, patient-derived human endothelial cells and peripheral blood mononuclear cells (PBMCs) were treated with mimics of bacterial infection and iron injury, before short-read RNA sequencing.

Results: 93 patients were identified with PAVMs- 19 had migraine, 8 had brain abscess. No variants reached genome-wide significance for migraine. AVT revealed a significant association between TJP3 loss-of-function gene variants and migraine (p=0.005), with minor allele frequencies 0.19 (cases), 0.02 (controls, p=0.001). For brain abscess, SAIGE-GENE+ indicated potential associations with loss-of-function variants in TJP3, NOD2, and SELE. All three were expressed in PBMCs, two in endothelial cells, with no change in THP3 or NOD2 expression after but increased endothelial SELE expression after both treatments.

Discussion: This study identifies a candidate gene for migraine susceptibility and suggests potential protective effects of loss-of-function TJP3 and NOD2 variants against brain abscess.

Leiomyosarcoma (LMS) is a smooth muscle malignancy that annually impacts more than 500 people in the UK. Surgery and chemotherapy remain the mainstays of treatment; however, recurrence rates reach ~70%, highlighting the need for novel therapeutic strategies. Genomic analyses have identified homologous recombination deficiency (HRD) in a subset of LMS, raising the potential for DNA damage response (DDR) agents such as poly-ADP ribose polymerase (PARP) inhibitors. Identifying patient subgroups who would benefit from DDR therapies remains a significant challenge.

This study aims to identify biomarkers predicting DDR agent response in LMS. Whole genome sequencing analysis of clinical LMS samples from the Pan-Cancer Analysis of Whole Genomes study (n=15) and Genomics England Research Environment (n=84) revealed low frequencies of somatic single nucleotide variants and small insertions and deletions, but high frequencies of copy number and structural variants in DDR genes. Using a genomic scar-based HRD prediction model, CHORD, BRCA2-type HRD was observed in 1/15 PCAWG and 1/84 GEL samples. Drug sensitivity experiments across four LMS cell lines demonstrated promising sensitivity to CHK1, ATR, Wee1 and ATM inhibitors. RAD51 immunofluorescence revealed variable foci formation following DNA damage induction across these lines.

Future work will involve genomic profiling of LMS in vitro models and evaluate how they recapitulate genomic changes observed in clinical samples. Integrating genomic analysis with in vitro drug sensitivity data aims to identify candidate predictive biomarkers to guide DDR-targeted therapy in LMS and improve patient outcomes.

Alport syndrome research has focused on severe X-linked (1:5,000) and autosomal recessive (1:40,000) forms, while the most prevalent genetic state heterozygosity for pathogenic COL4A3/COL4A4 variants (population frequency ~1:106) remains poorly characterised. Historically labelled 'carriers', these individuals exhibit broad clinical spectra, yet population-level penetrance of haematuria and proteinuria remains unquantified due to ascertainment bias in clinical cohorts.

We used large-scale population biobanks to quantify disease burden attributable to rare COL4A3/A4 variant classes in unselected cohorts. Using UK Biobank whole-genome sequencing data (n=500k), we performed per-variant association analyses of haematuria and proteinuria, and grouped rare variants into three functional classes per gene: glycine substitutions, truncating, and NC1 domain missense variants. PheWAS was conducted using SAIGE across 1,962 ICD-10 phenotypes, with replication in the US All of Us cohort. A GWAS for glomerular haematuria was performed with and without adjustment for a weighted COL4A3/A4 burden score.

We identified 85 and 59 rare variants significantly associated with haematuria and proteinuria, respectively. Glycine substitutions showed equivalent effects in both genes. Truncating and NC1 variants demonstrated a strong gene-specific effect, with significant associations only in COL4A4, suggesting greater tolerance to these variant types in COL4A3. The GWAS of haematuria showed that the COL4A3/A4 locus was the dominant signal for glomerular haematuria, fully explained by rare variant effects, with no significant trans-acting modifiers identified.

These findings enable variant-specific risk stratification, inform clinical pathogenicity assertions, and refine the genetic architecture of collagen IV nephropathies.

To address the challenges of deploying a novel computationally-intensive method (in our case Nomaly), we had to implement two complementary approaches: a CUDA-based workflow on the UK Biobank Research Analysis Platform using GPU instances, and a multi-threaded C++ backend (via pybind11) within the Our Future Health environment. Together, these approaches delivered up to 75x speedup and cost reduction compared to a Python-based implementation.

The technology that we needed to solve this problem for, our Nomaly framework (Lu et al., Nature Communications, 2023) – unlike traditional associative methods – uses a biology knowledge-based ab initio approach for phenotype prediction from genomic data. The most expensive component of the core compute is the solution of hundreds of thousands of complex eigenproblems derived from functional distances between genotypes. This architecture presents a severe deployment challenge in research environments due to their costs, restrictive nature, and their design centred on traditional well-known methods.

Our results demonstrate that novel, computationally intensive methods can be effectively scaled to meet the memory and processing demands of whole-genome and whole-exome datasets, even within a trusted research environment. Building on our experience with biobank-scale analyses, we will extend our framework to Genomics England rare disease and cancer cohorts, aiming to discover novel genetic causes of diseases. This will be supported by large-scale deployment within the Genomics England CloudOS platform, utilising the newly released AggV3 genomic dataset and using GPU-accelerated workflows and cloud-optimised Workflow Description Language (WDL) pipelines.

Physical activity (PA) is an important lifestyle intervention to combat obesity, although individual results vary. This study aimed to correlate single-nucleotide polymorphisms (SNPs) with body mass index (BMI) and body fat percentage (BFP) to better understand the relationship between SNPs, PA and body composition (BC).An institutional ethics approval (ETH2324-3589) was obtained to recruit PA participants. A total of fifty-six participants aged 18-65 years with no underlying medical conditions were recruited. The International PA Questionnaire was used to classify participants into sedentary/light PA(n=14), moderate PA(n=11), and vigorous PA (VPA) groups(n=31). Venous blood samples were collected and BMI and BFP recorded for all the participants. Extracted DNA was genotyped for SNPs in FTO, MC4R genes (n=56) and Reduced representation bisulphite sequencing (RRBS, n=9) was performed on all the three groups. Results were statistically analysed for associations between genotypes, BC and PA. Participants were grouped for FTO risk allele A (AA/AT, n=42) or wildtype (TT, n=13), MC4R risk allele C (CC/CT, n=25) or wildtype (TT n=31). Correlation analysis revealed that FTO SNP rs9939609 ‘A’ risk allele had a significant negative association with BFP in the VPA group (P=0.0457); MC4R SNP rs17782313 ‘C’ risk allele had a significant positive association with BMI in the VPA group (P=0.0466). RRBS DNA methylation levels data was compared within three PA groups which revealed that differentially methylated regions belonged to the 12 genes linked to obesity. The pilot data from this study concludes that genotypes and differentially methylated regions can predict positive molecular effects of PA.

Physical activity (PA) is an important lifestyle intervention to combat obesity, although individual results vary. This study aimed to correlate single-nucleotide polymorphisms (SNPs) with body mass index (BMI) and body fat percentage (BFP) to better understand the relationship between SNPs, PA and body composition (BC).An institutional ethics approval (ETH2324-3589) was obtained to recruit PA participants. A total of fifty-six participants aged 18-65 years with no underlying medical conditions were recruited. The International PA Questionnaire was used to classify participants into sedentary/light PA(n=14), moderate PA(n=11), and vigorous PA (VPA) groups(n=31). Venous blood samples were collected and BMI and BFP recorded for all the participants. Extracted DNA was genotyped for SNPs in FTO, MC4R genes (n=56) and Reduced representation bisulphite sequencing (RRBS, n=9) was performed on all the three groups. Results were statistically analysed for associations between genotypes, BC and PA. Participants were grouped for FTO risk allele A (AA/AT, n=42) or wildtype (TT, n=13), MC4R risk allele C (CC/CT, n=25) or wildtype (TT n=31). Correlation analysis revealed that FTO SNP rs9939609 ‘A’ risk allele had a significant negative association with BFP in the VPA group (P=0.0457); MC4R SNP rs17782313 ‘C’ risk allele had a significant positive association with BMI in the VPA group (P=0.0466). RRBS DNA methylation levels data was compared within three PA groups which revealed that differentially methylated regions belonged to the 12 genes linked to obesity. The pilot data from this study concludes that genotypes and differentially methylated regions can predict positive molecular effects of PA.

Disruption of gene functions essential to early eye development can lead to microphthalmia, anophthalmia, or coloboma (MAC); characterised by underdeveloped, absent, or fissured eyes, respectively. MAC are clinically and genetically heterogeneous, accounting for ~20% of childhood visual impairment worldwide. Despite over 100 causative genes, >70% of patients remain without a molecular diagnosis, highlighting a need to identify pathogenic variants beyond protein-coding regions.

To address this low diagnostic yield, we analyzed the genomes of 440 MAC families from the 100,000 Genomes Project to identify rare Structural Variants (SVs) intersecting 160 zebrafish-derived Highly Conserved Regions (HCRs). We applied a bioinformatics pipeline leveraging evolutionary conservation to identify HCRs as functional proxies for human enhancers at key developmental loci. This approach, alongside pedigree-aware filtering and IGV validation, identified five high-confidence candidate regulatory SVs across five probands. Two variants disrupted conserved elements at established MAC genes (HMX1,SALL1), while three implicated novel genomic contexts.

Regulatory landscape analysis revealed convergent evidence of enhancer disruption, MAC-relevant transcription factor occupancy, and topological domain architecture. Varied inheritance patterns were observed, consistent with proposed oligogenic architecture of MAC. To functionally validate these candidates, future work will leverage CRISPR-mediated deletion of orthologous HCRs in zebrafish and single-cell transcriptomic and ATAC-seq profiling of early optic vesicle stages, establishing causal links between regulatory disruption and MAC-relevant expression defects.

Overall, this study investigated non-coding regulatory variation as a potential mechanism in MAC, offering a framework to resolve molecular diagnoses in cases lacking a genetic explanation and advancing our understanding of the aetiology of MAC conditions.

Neurodevelopmental disorders (NDDs) are genetically heterogeneous conditions with a substantial contribution from rare deleterious variants. However, individuals of non-European ancestry remain underrepresented in large-scale genomic studies, limiting the generalisability of current findings. This study aims to characterise the genetic architecture of NDDs in South and East Asian populations using the 100,000 Genomes Project, with a particular focus on rare deleterious and homozygous loss-of-function variants enriched through autozygosity.

I applied a gene-based burden framework focusing on high-confidence predicted loss-of-function (pLOF) variants and deleterious missense variants. Variants were filtered for rarity and predicted functional impact, then collapsed at the gene level to quantify carrier counts per gene. Differences in carrier frequencies between Asian and non-Asian NDD cases were assessed using Fisher’s exact test to identify population-specific enrichments.

This approach provides a computationally efficient and interpretable strategy for detecting differences in rare variant burden across populations in large-scale sequencing datasets. The results highlight candidate genes with differential burden patterns, suggesting population-specific contributions to the genetic architecture of NDDs. These findings likely reflect underlying demographic processes, including consanguinity and fine-scale population structure, which shape the distribution of rare deleterious variation.

By leveraging diverse genomic data within the 100kGP, this study contributes to improving representation in genomic research and provides a foundation for downstream analyses, including replication, functional prioritisation, and extension to complementary cohorts such as Genes & Health

Main challenge in rare-disease genomics is to explain why complete loss of some genes is tolerated, whereas loss of others causes severe developmental disorders. Paralogs, duplicated genes within the same genome, offer an explanation because one copy may compensate for loss of the other. Yet the presence of a paralog alone does not fully explain why some gene losses are tolerated whereas others cause disease: more than 60% of human protein-coding genes have at least one paralog, but so do nearly 80% of Mendelian disease genes. Although paralog compensation has been demonstrated in yeast and cancer cell lines, its determinants in humans remain unclear. Here, we analysed four population-based sequencing cohorts totalling more than 1.1 million individuals and identified more than 2,000 genes with homozygous loss-of-function variants in healthy individuals, termed knockout (KO) genes. Paralogs were enriched among KO genes (OR=1.47, P=2.1×10−9), supporting compensation in humans. We then compared paralog KO genes with Gene2Phenotype paralog genes for which loss causes developmental disorder. Tolerance varied according to duplication origin. Among genes retained from ancient whole-genome duplication (WGD), greater sequence identity to the closest paralog was associated with a lower probability of being observed as a KO (OR=0.74, P=4.07×10−11), and highly similar WGD pairs were over-represented among developmental-disorder genes. This pattern was not observed among small-scale duplicates (SSDs). Together, these findings identify duplication origin as a key determinant of paralog compensation and provide an evolutionary framework for rare-disease genomics, helping explain when gene loss is tolerated and when it causes developmental disorders.

Main challenge in rare-disease genomics is to explain why complete loss of some genes is tolerated, whereas loss of others causes severe developmental disorders. Paralogs, duplicated genes within the same genome, offer an explanation because one copy may compensate for loss of the other. Yet the presence of a paralog alone does not fully explain why some gene losses are tolerated whereas others cause disease: more than 60% of human protein-coding genes have at least one paralog, but so do nearly 80% of Mendelian disease genes. Although paralog compensation has been demonstrated in yeast and cancer cell lines, its determinants in humans remain unclear. Here, we analysed four population-based sequencing cohorts totalling more than 1.1 million individuals and identified more than 2,000 genes with homozygous loss-of-function variants in healthy individuals, termed knockout (KO) genes. Paralogs were enriched among KO genes (OR=1.47, P=2.1×10−9), supporting compensation in humans. We then compared paralog KO genes with Gene2Phenotype paralog genes for which loss causes developmental disorder. Tolerance varied according to duplication origin. Among genes retained from ancient whole-genome duplication (WGD), greater sequence identity to the closest paralog was associated with a lower probability of being observed as a KO (OR=0.74, P=4.07×10−11), and highly similar WGD pairs were over-represented among developmental-disorder genes. This pattern was not observed among small-scale duplicates (SSDs). Together, these findings identify duplication origin as a key determinant of paralog compensation and provide an evolutionary framework for rare-disease genomics, helping explain when gene loss is tolerated and when it causes developmental disorders.

Research into the genetic basis of cancer has largely overlooked variation on repetitive regions of the genome, mainly due to associated technical challenges. Recent advances in sequencing and analytical methods have brought along exciting new perspectives to the field. For instance, in some cancers, copy number variation (CNV) has been reported in ribosomal DNA (rDNA), the multi-locus, multi-copy sequence that encodes the RNA components of ribosomes. However, the exact nature and consequences of these changes are still not completely understood. In this project, the largest of its kind to date, we are leveraging matched germline and tumour whole-genome sequencing (WGS) data from 11,792 Genomics England participants to characterise rDNA CNV in a large-scale, longitudinal pan-cancer cohort. We find that rDNA CNV is widespread in tumours, with 66% of samples showing overall rDNA CN loss. Using the corresponding genetic and phenotypic data available in Genomics England, we aim to assess the potential causes, consequences and clinical relevance of rDNA CNV. For instance, by attempting to disentangle the possible upstream processes resulting in rDNA CNV, and testing associations with clinical parameters (e.g., tumour stage, survival…). We believe that elucidating the phenotypic impact of rDNA alterations can help advance our understanding of genetic alteration on repetitive genomic regions, their impact on tumour biology and evolution, and possibly inform translational applications.

Whole genome sequencing (WGS) is currently the standard of care in the NHS for certain indications, while others still rely on exome sequencing coupled with gene panel approaches. Unfortunately, diagnostic rates for WGS remain at around 25%, which can partially be attributed to limitations in detecting pathogenic variants in non-coding regions of the genome.

The DeCODED (Determining the Causative genetic variants in individuals with Only a clinical Diagnosis of gEnetic Disease) Study aims to address some of the limitations of current WGS diagnostic pathways. Probands suspected of having rare, specific monogenic conditions in whom earlier WGS did not identify a diagnostic finding, are recruited after rigorous selection on clinical criteria, re-phenotyped, and pathogenic variant screening, including regulatory regions and utilising optical genome mapping and nanopore-based long-read sequencing, is carried out. To date, 17 probands have been enrolled in the study, with pathogenic non-coding variants identified in four cases and allowing a definitive genetic diagnosis to be returned to the affected families.

As an illustration, we present a case of adenylosuccinate lyase (ADSL) deficiency in which a genetic diagnosis was uncovered. Analysis of WGS data, accessed through the National Genomic Research Library, allowed the identification of a deep intronic variant predicted to create a cryptic exon in the transcript of the ADSL gene by cutting-edge splicing prediction tools. These predictions were later validated experimentally. Our work has allowed the affected family to finally receive a definitive genetic diagnosis, enabling prenatal or pre-implantation genetic testing. 

Brittle cornea syndrome (BSC) is a rare, autosomal recessive connective tissue disorder, primarily characterised by severe corneal thinning and potential rupture. Variants in PRDM5 or ZNF469 are the only known causes to date. BCS was confirmed as the clinical diagnosis for a female patient seen by the North West Thames Regional Genetics Service, whose medical history includes non-traumatic right and left corneal rupture, alongside blue sclerae, joint hypermobility and hearing loss. Genetic studies identified a heterozygous maternally inherited missense variant in PRDM5, reported as a Tier 3 variant of uncertain significance (VUS) following Genomics England Criteria. The patient was recruited into the 100,000 Genomes study for whole genome sequencing (WGS), however no second coding variant was identified. Subsequent inclusion in the NEEDS (Natural history Exploration of rare EDS types) study led to further analysis for a potential non-coding variant disrupting the second PRDM5 allele. Long-read trio WGS analysis led to the identification of an ultra-rare paternally inherited intronic variant located ~600 bases from the nearest exon boundary, strongly predicted by SpliceAI to create a cryptic exon by inducing donor and acceptor splice sites. Induced pluripotent stem cells (iPSCs) were derived from fibroblasts isolated from the proband and father and were analysed by RNA-seq (n=3 clones/donor). Both individuals carry the rare intronic variant and both showed RNA-seq read supporting the presence of the cryptic exon. Both variants have since been upgraded to pathogenic / likely pathogenic following ACGS guidelines and the findings have been reported to the referring clinical team.

Epilepsy is a complex neurological disorder that can be influenced by multiple genetic and environmental factors and is often associated with other health conditions. Identifying the comorbidities associated with epilepsy is important because they can affect outcomes and treatment choices. For example, medications that harm cognition should be avoided in people with significant cognitive dysfunction. This project focuses on estimating the Polygenic Risk Score (PRS) for various comorbidities. The PRS provides an estimate of an individual’s genetic liability to or trait or specific morbidity. Here, the PRS for intelligence, longevity, anxiety, dementia, migraine, heart disease, BMI, arthritis, and other phenotypes were estimated for 772 epilepsy cases and compared with 1187 healthy controls from the 100,000 genomes project, providing an assessment of the integrity of the genetic background of individuals with epilepsy. The results of this project have the potential to improve our understanding of the genetic basis of epilepsy and its comorbidities, and to facilitate the development of personalized approaches to epilepsy management and treatment.

Cerebrovascular malformations, such as Moyamoya disease (MMD) and Vein of Galen malformations (VOGM), are rare disorders impacting the structure and function of the cerebral vasculature; however, the full genetic complement remains elusive. Owing to their rarity, when the 100,000 Genomes Project was established, participants with either MMD or a VOGM were actively recruited to the study. This project sought to use the existing scientific literature to update the gene list on R336 on PanelApp and apply the updated list to a study cohort of 100,000 Genomes study participants with MMD or VOGM. The literature search led to the addition of NOS3, DIAPH1 and ANO1 onto the R336 PanelApp gene list, and the update of pre-existing genes EPHB4 and CBL. The gene list was applied to a study cohort created within the Genomics England Research Environment. The study cohort constituted mostly young adult females with MMD born after the year 2000, broadly representing the MMD population which exhibits a female preponderance. There were fewer participants with VOGM, highlighting its rarity in the population. Most variants identified in the study were identified in the C-terminal region of RNF213, a gene predominantly associated with MMD. Of the variants identified, 3 were classified as pathogenic and 3 were variants of uncertain signifcance. Furthermore, potential disease-causing variants were also identified in NOS3 and CBL. Overall, the study increased the yield of diagnoses in the cerebrovascular malformation 100,000 Genomes population and highlights the utility of periodic data reanalysis using updates from the scientific literature.

Background: Genome-wide association studies (GWAS) identified genetic loci for complex traits using the traditional univariate approach. The multivariate methods, including “reverse regression”-based SCOPA software tool, are more powerful over univariate models to detect multi-phenotype effects. To date, SCOPA, supported text-based formats only, limiting it from application on biobank-scale data. We extended SCOPA to a computationally efficient for large-scale data and evaluated its performance in a UK biobank-based multi-phenotype GWAS.

Material and Methods: The extended version of SCOPA utilises BGEN in addition to GEN input files, while preserving all original features, including accommodation of both directly genotyped and imputed variants and model selection implementing the Bayesian Information Criterion (BIC). The performance of the BGEN-compatible version of SCOPA was evaluated in a multi-phenotype GWAS on type 2 diabetes (T2D) and colorectal cancer (CRC) in 487,409 UKBB individuals.

Results: We identified 73 independent genome-wide significant signals at 68 distinct loci jointly associated with T2D and CRC (two-phenotype model P-value<5×10-8). Of these, 46 loci have previously been associated with T2D, ten with CRC and 11 with both diseases. Among the 11 genetic loci previously associated with both diseases, the phenotype subset with minimum BIC included only T2D for 10 loci (IGF2BP2, RREB1, ANK1, SLC30A8, CDKN2B-AS1, ZMIZ1, TCF7L2, CCND2, RETREG3 and CEBPB/SMIM25) whereas only PNKD was best explained by CRC alone.

Conclusion: The multi-phenotype GWAS in large datasets enables dissection of genetic effects at individual loci for correlated phenotypes. We illustrated the utility of the BGEN-format implementation in SCOPA to facilitate such analyses."

Protein lipidation is defined as ‘enzymatic processes that act directly on a protein substrate to covalently attach a lipid modification.' This project investigated the phenotypic severity within monogenic disorders of protein lipidation using data from the 100,000 Genomes project. A cohort of 231 individuals was assembled, recruited for undiagnosed rare disease presentations, carrying tiered variants in nine genes involved in lipidation pathways. A novel domain-based scoring framework was developed, derived from Human Phenotype Ontology classification, to quantify phenotypic severity and breadth across 17 organ systems. Gene-level predictors of these phenotypic severity metrics were then analysed.

Clinical presentations were highly heterogeneous, with a wide range of severities. Statistical analysis showed that significant differences in severity exist between lipidation disorders, influenced by the lipidation pathway involved; palmitoylation disorders revealed higher average severity and breadth scores compared to Palmitoleoylation, although the effect size was modest. Heatmaps highlighted the nervous system and musculature as the main drivers of phenotypic severity in palmitoylation and GPI-anchoring pathways. The analysis revealed that males and individuals with an X-linked recessive disorder, in contrast to X-linked dominant disorders, experienced higher phenotypic severity scores. Although missense constraint metrics did not show a clear monotonic relationship with phenotypic severity, a protein's functional interaction network did show a moderately strong positive correlation with severity.

These findings suggest that biological properties of specific lipidation pathways, such as differential tissue expression and functional roles in neuronal processes, could contribute to clinical variation in severity, highlighting areas of targeted therapeutic research in rare lipidation disorders.

Background: Missense variants with low predicted missense pathogenicity scores are regularly discarded during diagnostic variant assessment. However, missense variants can disrupt splicing through exon skipping, loss of exonic splicing regulatory elements, or the activation of cryptic splice sites, which may be overlooked. We sought to quantify the overall proportion of spliceogenic missense variants and identify novel pathogenic variants.

Methods: We performed a large-scale analysis of the splicing potential of 8.3 million missense variants in MANE Select transcripts from UK Biobank (UKB) using SpliceAI, REVEL and AlphaMissense. Downstream consequences were assessed using circulating levels of 963 plasma proteins. We subsequently investigated likely spliceogenic missense variants in the 100,000 Genomes Project (100kGP).

Results: We found 225,430 (2.7%) missense variants in UKB with a SpliceAI score indicative of splice disruption (Δ>0.2), of which 162,528 (72%) also had low missense pathogenicity scores (REVEL<0.5 or AlphaMissense<0.34). Despite having low missense scores, these variants were associated with decreased circulating protein levels (ꞵ=-0.61) comparable to loss-of-function variants (ꞵ=-0.94) and significantly different from tolerated missense variants (ꞵ=-0.05). In 100kGP, we identified three rare missense variants in four unsolved patients, with strong evidence that aberrant splicing is the disease mechanism. Additionally, we provide strong evidence to reclassify four rare missense variants of uncertain significance as likely pathogenic.

Conclusion: Our findings suggest that ~2-3% of missense variants may affect splicing, supporting the use of splice prediction tools for diagnostic assessment of missense variants. "

Disruption of genes essential to early eye development can lead to anophthalmia or microphthalmia (A/M) which are characterised by absent or underdeveloped eyes, respectively. A/M are clinically and genetically heterogeneous, with their aetiology still not fully understood. This results in a low molecular diagnostic rate (~20–30%), limiting clinical management and genetic counselling. Furthermore, pathogenic variants may remain undetected due to incomplete gene panels and variant interpretation deficiencies. This study aimed to increase the yield of A/M pathogenic variant identification by implementing an enhanced variant-investigation framework.

We curated an updated A/M gene panel of 141 genes through a systematic literature review, and screened for rare loss-of-function, missense, splicing, and structural variants, which were annotated using deep-learning tools (AlphaMissense, SpliceAI) and other in silico predictors (REVEL, Missense3D, and molecular dynamics simulations). We identified candidate variants in 35 probands, increasing the candidate detection rate from 15.2% (43/283) to 27.6% (78/283). Across the 141 A/M-associated genes in our panel, variants were most frequently affected were identified in established A/M genes but also highlighted emerging candidates, including ACTG1, CAMK2B, DYRK1A, HDAC6, NR2F1, RERE, SIX3, and TRAF7, strengthening their involvement in A/M pathogenesis.

This study increased the yield of A/M pathogenic variants and strengthened genotype-phenotype correlations by implementing an enhanced variant-investigation framework integrating a curated gene panel, comprehensive variant detection pipelines, and advanced in silico analysis. Nonetheless, the modest diagnostic rate highlights the genetic complexity of A/M and the ongoing need for improved functional annotation and discovery of additional disease-relevant genes.

Fuchs endothelial corneal dystrophy (FECD) is an age-related inherited corneal disease. It is genetically heterogeneous, with up to 81% of patients of European ancestry harboring expansions (≥50 copies) of a CTG repeat in an intronic region of the TCF4 gene. Additional genetic causes have been identified including rare variants in COL8A2, MIR184, SLC4A11, and ZEB1. However, our recent genetic analysis of a TCF4 expansion-negative (Exp-) FECD cohort highlights that these rare variants account for a minority of cases, leaving the remaining 96% unresolved.

This study aims to investigate this missing heritability, by deploying the Exomiser-based analytical framework, geneBurdenRD, previously applied to the rare disease data of the 100,000 Genomes Project (100kGP) (Cipriani et al., 2025). We compared the burden of Exomiser candidate rare protein-coding variants from whole-genome sequencing in 122 unresolved Exp- FECD cases versus 10,080 single proband participants affected by non-ophthalmological rare diseases recruited to the 100kGP as ‘controls’.

A total of 87 new associations were identified (false discovery rate <5%), including 14 with evidence of expression in the corneal endothelium (TPM >10). In silico triaging revealed three promising associations with strongest overall genetic and experimental evidence, RBM25, SEC24A, and SPTLC1, involved in pre-mRNA splicing, protein transport, and sphingolipid biosynthesis, respectively.

Further functional analysis in corneal cell models and segregation analysis in family members could validate these candidates and increase the diagnostic yield by over 15%, highlighting the clinical potential of large-scale statistical approaches for rare disease-gene association discovery.

Individuals with intellectual disability and psychiatric comorbidities (ID+) have been reported to have higher rates of copy number variants (CNVs) within neurodevelopmental risk loci than individuals with only intellectual disability (ID). Furthermore, the impact of a co-occurring functional single-nucleotide variant (SNV) in the opposite strand to a deletion CNV hasn’t been explored within the ID cohorts.

We performed one of the largest analyses to ascertain the NDD CNV rates in participants with ID from the 100,000 Genomics Project, utilising clinical phenotypic data from electronic health records. CNVs were extracted from 63 NDD risk loci and co-occurring SNVs from the opposite strand of deleted CNVs using whole-genome sequence data.
Overall, the NDD CNV rate was 2.8%, which was higher than that of healthy controls in other studies; however, this rate was lower than that of previous ID-specific studies.

In contrast to previous studies, we found that the NDD CNV rate was higher in the ID-only group (3.9%) than in ID+ (2.5%). Notably, 16p13.11 duplication had the highest frequency of CNVs. Furthermore, 46.8% of deleted CNVs harboured co-occurring deleterious SNVs in the opposite strand. Recurrent double hit identified 6 participants involving the 16p11.2 locus and deleterious missense variant in QPRT - a well-established gene linked to autism susceptibility.

This appears to be the first study to explore CNV rates in NDD risk loci in a larger cohort, and compound heterozygosity in individuals ID/ID+. This highlights the significance of using WGS-based approaches to assess the impact of variants in this population.

Genomic studies of bleeding and platelet disorders (BPD) have focused on European outbred populations, limiting discovery of recessive disease and ancestry-restricted variants. Populations with frequent parental relatedness provide a model for identifying rare homozygous variants and genotype–phenotype associations.

We assessed associations between bleeding phenotypes and genomic autozygosity in a UK population enriched for consanguinity and characterised rare homozygous variation in BPD genes. Whole exome and health record data were analysed from 44,190 British Bangladeshi and Pakistani participants in the Genes & Health study. Bleeding phenotypes were derived from ICD-10 codes and summarised as a bleeding score (ICD-BAT). Autozygosity was quantified using runs of homozygosity (FROH). Rare coding variants (MAF <1%) in 116 R90 BPD genes were identified.

Participants were 45% male, with mean age 41 years (range 15–101). Mean FROH was 0.0168, with 16.8% showing high consanguinity (FROH >0.0625). Median ICD-BAT score was 0 (range 0–4). Higher scores were associated with increasing autozygosity (β = 0.20 for FROH = 0.0625; p = 0.005).

Overall, 1,764 participants (4.0%) harboured 808 rare homozygous missense variants across 98 BPD genes. Sixteen (0.04%) carried 12 rare homozygous loss-of-function (LoF) variants. Only two were reported in homozygosity in gnomAD (n = 807,162), both in South Asian individuals. Probability of homozygous LoF variants increased with autozygosity, reaching 0.26% for FROH >0.0625.

These findings show enrichment of rare homozygous variants in BPD genes associated with autozygosity and bleeding phenotypes. Most variants are absent from European reference datasets, highlighting the importance of diverse populations for recessive disease discovery and equitable genomic medicine."

Despite whole genome sequencing (WGS) being available to thousands of patients with rare disease, the majority remain without a molecular diagnosis. Much of this missing heritability is likely explained by variants that are difficult to interpret — including structural variants (SVs) that delete gene regulatory elements known as enhancers. Identifying such deletions has historically been challenging because the relationships between enhancers and the genes they control are poorly characterised at the resolution needed for clinical interpretation.

To address this, we performed Micro Capture-C (MCC) — an assay that detects physical contacts between regulatory elements at base-pair resolution — in primary adult human neurons and oligodendrocytes, generating precise maps linking enhancers to their target genes across the genome. We applied a coverage-based filtering approach to Manta SV calls from WGS data across the Genomics England rare disease case cohort (n=58,548). By requiring absence of read coverage across deletion sites, we identified homozygous and hemizygous deletions overlapping MCC-defined enhancers of key neurodevelopmental genes, restricting analysis to cases with neurodevelopmental diagnoses.

This approach identified 16 individuals with a neurodevelopmental disorder — principally intellectual disability — carrying a candidate pathogenic enhancer deletion. This may represent an underestimate of prevalence given the constraints of the current enhancer map.

A key limitation is that our experimental data derive from adult brain. We have therefore trained and validated a machine learning model to predict MCC interaction profiles from widely available genomic signals, which we are applying to foetal brain data to improve diagnostic sensitivity in a developmental context.

RNA sequencing datasets are used widely to explore gene expression, yet the optimal method by which to analyse the data remains elusive, with existing methods often generating different results.

Twelve RNAseq datasets from GEO database for raw RNAseq counts from Parkinson’s disease (PD) cases and controls (brain, blood, iPSC, and skeletal muscle) were normalised using median ratio normalisation. A negative binomial generalised linear model (glm.nb) was used to test for differential expression. Given substantial heterogeneity across datasets and inconsistent availability of study-specific covariates, we incorporated the first two principal components (PC1 and PC2) derived from genome-wide expression to model latent technical variation. Lambda values and qq-plots were compared against an empirical null distribution for each dataset. Differential expression was also assessed using DESeq2 and Limma.

Inclusion of PC1 and PC2 substantially reduced heterogeneity across datasets and improved calibration (λ ≈ 1) for all methods. Under the null, our method demonstrated stable control of p-value inflation across studies of varying sample size. Under the alternative, our approach outperformed DESeq2 and limma in detecting 45 established PD genes from Genomics England. This improvement was evident at both α = 0.05 and after Bonferroni correction (α’ = 0.001), with the largest gain observed under stringent significance thresholds.

Heterogeneity in public transcriptomic datasets is a major challenge, particularly when metadata are frequently incomplete. By explicitly modelling latent technical variation through genome-wide principal components, our approach improves calibration and power, providing a reliable and scalable framework for differential expression analysis in complex disease studies.

Around half of all patients with rare neurological disorders do not receive a genetic diagnosis, despite undergoing whole genome sequencing. Most research focuses on the ~1% of the human genome containing protein-coding genes. Genetic variants which fall outside of protein-coding regions are increasingly being recognised for their role in rare disorders, but are not routinely examined in rare disease diagnostics. These non-coding regions play important roles in regulating when and where genes are expressed, but can be difficult to study. Availability of large-scale sequencing data from projects including the 100,000 Genomes Project and UK Biobank offers a relatively untapped opportunity to explore and increase our understanding of the non-coding genome. I use these data to identify non-coding sites that are resistant to genetic variation and therefore likely to be functionally important. This is accomplished using disease-agnostic bioinformatic and statistical approaches such as the Mutability Adjusted Proportion of Singletons (MAPS), which can identify regions of the genome that are under selective constraint and therefore lack variation in the general population. I investigate an extensive set of non-coding regions, including the untranslated regions of genes, introns, promoters, and other regulatory elements such as transcription factor binding sites. This work will improve our understanding of the function of the non-coding genome, and identify conserved non-coding regions of potential disease relevance. Candidate functional regions will be assessed in individuals with rare disease, to enable the discovery of rare, potentially pathogenic variants.

Dr Claudia P. Cabrera is a Senior Lecturer in Bioinformatics at the William Harvey Research Institute, Queen Mary University of London. Her research integrates large‑scale genomic data, clinical phenotypes, and computational approaches to advance our understanding of pharmacogenomics and complex traits. She is interested in uncovering clinically actionable pharmacogenomic variants that can enhance drug safety and support fair, effective precision‑medicine strategies for individuals from diverse populations.

Postdoctoral researcher in machine learning and genomics, developing sequence-based transcriptomic models to interpret non-coding variants and improve rare disease diagnosis.

I am a King’s Health Partners Postdoctoral Clinical Research Fellow at King’s College London and a consultant in Clinical Genetics at Guy’s and St Thomas’ Hospital. My work integrates neuroimaging and bioinformatics to investigate the genetic architecture of brain development, neurodegeneration, and related cognitive, neurodevelopmental, and psychiatric phenotypes. 

Dr. Karen Low is a consultant clinical geneticist based in Bristol with a special interest in genetic conditions associated with developmental delay, intellectual disability, and autism. She has conducted research on several rare syndromes, including PUF60-related syndrome and HUWE1-related syndrome, and has particular expertise in KBG syndrome. 
 
Dr. Low serves on the scientific advisory board of the KBG Foundation, organized a UK KBG Family Day in partnership with Unique, and has authored four patient information leaflets for families. She has led a national study in the UK of over 500 children with genetic neurodevelopmental conditions – GenROC. Through her clinical and research work, she has seen firsthand the challenges families face when limited information is available about their child’s genetic condition—particularly how knowledge gaps can affect clinical care. She is passionate about involving patients in research.  Her PhD focuses on learning from parent-reported data in genetic neurodevelopmental conditions.  

Caroline is a first year clinical PhD fellow and ST3 histopathology resident doctor, based in Leeds and funded by Cancer Research UK through Manchester Cancer Research Centre. Her project is based on investigating whether the cancer-associated microbiome is a key determinant of the immune response to bowel cancer or if immunogenetic lesions are more important.

Postdoctoral researcher in machine learning and genomics, developing sequence-based transcriptomic models to interpret non-coding variants and improve rare disease diagnosis.

Katrina Andrews is a Clical PhD student at the Wellcome Sanger Institute in the Hurles lab and a resident doctor in Clinical Genetics at Addenbrooke's Hospital, Cambridge. 

Her research focuses on improving diagnostic yields in developmental disorders, with a particular interest in altered-function (e.g. gain-of-function) disease mechanisms - a class of predominantly missense-driven mutations poorly detected by current tools. Her work includes characterising altered-function genes and variants, and leveraging somatic mutation data from cancer and healthy tissues as a clinically actionable evidence source for germline variant

Gabriel Funingana is a medical oncologist and clinical academic focused on advancing precision cancer medicine through molecular oncology, biomarker development, and the clinical implementation of genomic technologies. He is a Clinical Research Training Fellow and PhD student at the University of Cambridge and Cambridge University Hospitals NHS Foundation Trust, supported by The Mark Foundation Institute for Integrative Cancer Medicine.

Gabriel’s work sits at the interface of pan-cancer genomics, translational oncology, and clinical decision-making. He has contributed to first-in-human and first-in-class clinical trials, as well as personalised medicine studies designed to match patients to therapies based on the molecular characteristics of their tumours. His current research explores how whole-genome sequencing and AI-enabled decision-support tools can be embedded into clinical pathways to improve treatment selection and patient outcomes.

Nouman Ahmed is a Data and AI Engineer at the University of Oxford’s Nuffield Department of Women’s & Reproductive Health, working with the Deep Medicine group. His research sits at the intersection of artificial intelligence, large-scale health data, and clinical risk prediction, with experience developing machine learning models using electronic health records across multi-cohort settings. Nouman’s current and emerging work focuses on applying AI and big data methods to improve early identification, risk stratification, and outcomes in women’s health, including polycystic ovary syndrome and endometriosis. He will be starting his PhD in October with the same group.

Emily is a ST3 NIHR Academic Clinical Fellow in Clinical Genetics at Sheffield Children's Hospital and a Bicentennial Postdoctoral Research Fellow at the University of Sheffield. Her research focuses on rare genomic disease and novel therapeutic strategies. She is a Paediatrician by background and completed MRCPCH in 2016. Her PhD in molecular genetics at Queen Mary University of London investigated novel genetic causes of growth failure in childhood, including the discovery of several non-coding variants causing classic disease phenotypes. 

Anthony is a DPhil candidate in Genomic Medicine and Statistics supervised by Nicky Whiffin and Jenny Taylor. His research focuses on using DNA structural variants, specifically biallelic deletions, to facilitate novel gene discovery and new diagnoses. He also has an interest in digital phenotyping using electronic healthcare record data.

Before his DPhil, Anthony worked as a statistician in the Civil Service, and graduated from the University of Cambridge in 2020 with an MPhil in Genomic Medicine and the University of Dundee in 2019 with a BSc in Biomedical Sciences.

Helen White is an experienced patient advocate and public involvement contributor with lived experience of endometrial cancer, later found to be due to Lynch syndrome. She served on the Participant Panel at Genomics England from 2018 to March 2026, including as Vice-Chair for Cancer (2023–2026), and was a patient representative for the former endometrial cancer GECIP. She is a member of the LynchVax PPI group and Trial Management Group, and leads the Peaches (Womb Cancer Trust) Patient Voices and Future Voices groups, connecting people with lived experience with researchers to help shape their research.