Predicting the impact of rare variants on RNA splicing in CAGI6.

Variants which disrupt splicing are a frequent cause of rare disease that have been under-ascertained clinically. Accurate and efficient methods to predict a variant's impact on splicing are needed to interpret the growing number of variants of unknown significance (VUS) identified by exome and genome sequencing. Here, we present the results of the CAGI6 Splicing VUS challenge, which invited predictions of the splicing impact of 56 variants ascertained clinically and functionally validated to determine splicing impact. The performance of 12 prediction methods, along with SpliceAI and CADD, was compared on the 56 functionally validated variants. The maximum accuracy achieved was 82% from two different approaches, one weighting SpliceAI scores by minor allele frequency, and one applying the recently published Splicing Prediction Pipeline (SPiP). SPiP performed optimally in terms of sensitivity, while an ensemble method combining multiple prediction tools and information from databases exceeded all others for specificity. Several challenge methods equalled or exceeded the performance of SpliceAI, with ultimate choice of prediction method likely to depend on experimental or clinical aims. One quarter of the variants were incorrectly predicted by at least 50% of the methods, highlighting the need for further improvements to splicing prediction methods for successful clinical application.

A gene pathogenicity tool "GenePy" identifies missed biallelic diagnoses in the 100,000 Genomes Project.

PURPOSE: The 100,000 Genomes Project diagnosed a quarter of affected participants, but 26% of diagnoses were not on the applied gene panel(s); with many being de novo variants. Assessing biallelic variants without a gene panel is more challenging. METHODS: We sought to identify missed biallelic diagnoses using GenePy, which incorporates allele frequency, zygosity, and a user-defined deleterious metric, generating an aggregate GenePy score per gene, per participant. We calculated GenePy scores for 2862 recessive disease genes in 78,216 100,000 Genomes Project participants. For each gene, we ranked participant GenePy scores and scrutinized affected participants without a diagnosis, whose scores ranked among the top 5 for each gene. In cases which participant phenotypes overlapped with the disease gene of interest, we extracted rare variants and applied phase, ClinVar, and ACMG classification. RESULTS: 3184 affected individuals without a molecular diagnosis had a top-5-ranked GenePy score and 682 of 3184 (21%) had phenotypes overlapping with a top-ranking gene. In 122 of 669 (18%) phenotype-matched cases (excluding 13 withdrawn participants), we identified a putative missed diagnosis (2.2% of all undiagnosed participants). A further 334 of 669 (50%) cases have a possible missed diagnosis but require functional validation. CONCLUSION: Applying GenePy at scale has identified 456 potential diagnoses, demonstrating the value of novel diagnostic strategies.

Markers of adipose tissue fibrogenesis associate with clinically significant liver fibrosis and are unchanged by synbiotic treatment in patients with NAFLD.

BACKGROUND AND AIMS: Subcutaneous adipose tissue (SAT) dysfunction contributes to NAFLD pathogenesis and may be influenced by the gut microbiota. Whether transcript profiles of SAT are associated with liver fibrosis and are influenced by synbiotic treatment (that changes the gut microbiome) is unknown. We investigated: (a) whether the presence of clinically significant, ≥F2 liver fibrosis associated with adipose tissue (AT) dysfunction, differential gene expression in SAT, and/or a marker of tissue fibrosis (Composite collagen gene expression (CCGE)); and (b) whether synbiotic treatment modified markers of AT dysfunction and the SAT transcriptome. METHODS: Sixty-two patients with NAFLD (60 % men) were studied before and after 12 months of treatment with synbiotic or placebo and provided SAT samples. Vibration-controlled transient elastography (VCTE)-validated thresholds were used to assess liver fibrosis. RNA-sequencing and histological analysis of SAT were performed to determine differential gene expression, CCGE and the presence of collagen fibres. Regression modelling and receiver operator characteristic curve analysis were used to test associations with, and risk prediction for, ≥F2 liver fibrosis. RESULTS: Patients with ≥F2 liver fibrosis (n = 24) had altered markers of AT dysfunction and a SAT gene expression signature characterised by enrichment of inflammatory and extracellular matrix-associated genes, compared to those with

Reduced penetrance of gene variants causing amyotrophic lateral sclerosis.

BACKGROUND: Amyotrophic lateral sclerosis overlaps aetiologically and genetically with frontotemporal dementia and occurs in both familial and apparently sporadic forms. The most commonly implicated genes are C9orf72, SOD1, TARDBP and FUS. Penetrance of disease-causing variants in these genes is known to be incomplete, but has not been well studied at population level. OBJECTIVE: We sought to determine the population-level penetrance of pathogenic and likely pathogenic variants in genes commonly causing amyotrophic lateral sclerosis. METHODS: Published epidemiological data for amyotrophic lateral sclerosis and frontotemporal dementia were used to calculate expected frequencies of disease-causing variants per gene at population level. Variant data from gnomAD and ClinVar databases were used to ascertain observed numbers of disease-causing variants and to estimate population-level penetrance per gene. Data for C9orf72 were obtained from the published literature. RESULTS: Maximum population penetrance for either amyotrophic lateral sclerosis or frontotemporal dementia was found to be 33% for C9orf72 (95% CI (20.9 to 53.2)), 54% for SOD1 (95% CI (32.7 to 88.6)), 38% for TARDBP (95% CI (21.1 to 69.8)) and 19% for FUS (95% CI (13.0 to 28.4)). CONCLUSION: Population-level penetrance of amyotrophic lateral sclerosis disease genes is reduced. This finding has implications for the genetic testing and counselling of affected individuals and their unaffected relatives.

A Panel-Agnostic Strategy 'HiPPo' Improves Diagnostic Efficiency in the UK Genomic Medicine Service.

Genome sequencing is available as a clinical test in the UK through the Genomic Medicine Service (GMS). The GMS analytical strategy predominantly filters genome data on preselected gene panels. Whilst this reduces variants requiring assessment by reporting laboratories, pathogenic variants outside applied panels may be missed, and variants in genes without established disease-gene relationships are largely ignored. This study compares the analysis of a research exome to a GMS clinical genome for the same patients. For the research exome, we applied a panel-agnostic approach filtering for variants with High Pathogenic Potential (HiPPo) using ClinVar, allele frequency, and in silico prediction tools. We then restricted HiPPo variants to Gene Curation Coalition (GenCC) disease genes. These results were compared with the GMS genome panel-based approach. Twenty-four participants from eight families underwent parallel research exome and GMS genome sequencing. Exome HiPPo analysis identified a similar number of variants as the GMS panel-based approach. GMS genome analysis returned two pathogenic variants and one de novo variant. Exome HiPPo analysis returned the same variants plus an additional pathogenic variant and three further de novo variants in novel genes, where case series are underway. When HiPPo was restricted to GenCC disease genes, statistically fewer variants required assessment to identify more pathogenic variants than reported by the GMS, giving a diagnostic rate per variant assessed of 20% for HiPPo versus 3% for the GMS. With UK plans to sequence 5 million genomes, strategies are needed to optimise genome analysis beyond gene panels whilst minimising the burden of variants requiring clinical assessment.

Systematic identification of disease-causing promoter and untranslated region variants in 8,040 undiagnosed individuals with rare disease.

Background: Both promoters and untranslated regions (UTRs) have critical regulatory roles, yet variants in these regions are largely excluded from clinical genetic testing due to difficulty in interpreting pathogenicity. The extent to which these regions may harbour diagnoses for individuals with rare disease is currently unknown. Methods: We present a framework for the identification and annotation of potentially deleterious proximal promoter and UTR variants in known dominant disease genes. We use this framework to annotate de novo variants (DNVs) in 8,040 undiagnosed individuals in the Genomics England 100,000 genomes project, which were subject to strict region-based filtering, clinical review, and validation studies where possible. In addition, we performed region and variant annotation-based burden testing in 7,862 unrelated probands against matched unaffected controls. Results: We prioritised eleven DNVs and identified an additional variant overlapping one of the eleven. Ten of these twelve variants (82%) are in genes that are a strong match to the individual's phenotype and six had not previously been identified. Through burden testing, we did not observe a significant enrichment of potentially deleterious promoter and/or UTR variants in individuals with rare disease collectively across any of our region or variant annotations. Conclusions: Overall, we demonstrate the value of screening promoters and UTRs to uncover additional diagnoses for previously undiagnosed individuals with rare disease and provide a framework for doing so without dramatically increasing interpretation burden.

PhenoScore quantifies phenotypic variation for rare genetic diseases by combining facial analysis with other clinical features using a machine-learning framework.

Several molecular and phenotypic algorithms exist that establish genotype-phenotype correlations, including facial recognition tools. However, no unified framework that investigates both facial data and other phenotypic data directly from individuals exists. We developed PhenoScore: an open-source, artificial intelligence-based phenomics framework, combining facial recognition technology with Human Phenotype Ontology data analysis to quantify phenotypic similarity. Here we show PhenoScore's ability to recognize distinct phenotypic entities by establishing recognizable phenotypes for 37 of 40 investigated syndromes against clinical features observed in individuals with other neurodevelopmental disorders and show it is an improvement on existing approaches. PhenoScore provides predictions for individuals with variants of unknown significance and enables sophisticated genotype-phenotype studies by testing hypotheses on possible phenotypic (sub)groups. PhenoScore confirmed previously known phenotypic subgroups caused by variants in the same gene for SATB1, SETBP1 and DEAF1 and provides objective clinical evidence for two distinct ADNP-related phenotypes, already established functionally.

A gene pathogenicity tool 'GenePy' identifies missed biallelic diagnoses in the 100,000 Genomes Project.

The 100,000 Genomes Project (100KGP) diagnosed a quarter of recruited affected participants, but 26% of diagnoses were in genes not on the chosen gene panel(s); with many being de novo variants of high impact. However, assessing biallelic variants without a gene panel is challenging, due to the number of variants requiring scrutiny. We sought to identify potential missed biallelic diagnoses independent of the gene panel applied using GenePy - a whole gene pathogenicity metric. GenePy scores all variants called in a given individual, incorporating allele frequency, zygosity, and a user-defined deleterious metric (CADD v1.6 applied herein). GenePy then combines all variant scores for individual genes, generating an aggregate score per gene, per participant. We calculated GenePy scores for 2862 recessive disease genes in 78,216 individuals in 100KGP. For each gene, we ranked participant GenePy scores for that gene, and scrutinised affected individuals without a diagnosis whose scores ranked amongst the top-5 for each gene. We assessed these participants' phenotypes for overlap with the disease gene associated phenotype for which they were highly ranked. Where phenotypes overlapped, we extracted rare variants in the gene of interest and applied phase, ClinVar and ACMG classification looking for putative causal biallelic variants. 3184 affected individuals without a molecular diagnosis had a top-5 ranked GenePy gene score and 682/3184 (21%) had phenotypes overlapping with one of the top-ranking genes. After removing 13 withdrawn participants, in 122/669 (18%) of the phenotype-matched cases, we identified a putative missed diagnosis in a top-ranked gene supported by phasing, ClinVar and ACMG classification. A further 334/669 (50%) of cases have a possible missed diagnosis but require functional validation. Applying GenePy at scale has identified potential diagnoses for 456/3183 (14%) of undiagnosed participants who had a top-5 ranked GenePy score in a recessive disease gene, whilst adding only 1.2 additional variants (per individual) for assessment.

Extending the phenotypes associated with TRIO gene variants in a cohort of 25 patients and review of the literature.

The TRIO gene encodes a rho guanine exchange factor, the function of which is to exchange GDP to GTP, and hence to activate Rho GTPases, and has been described to impact neurodevelopment. Specific genotype-to-phenotype correlations have been established previously describing striking differentiating features seen in variants located in specific domains of the TRIO gene that are associated with opposite effects on RAC1 activity. Currently, 32 cases with a TRIO gene alteration have been published in the medical literature. Here, we report an additional 25, previously unreported individuals who possess heterozygous TRIO variants and we review the literature. In addition, functional studies were performed on the c.4394A > G (N1465S) and c.6244-2A > G TRIO variants to provide evidence for their pathogenicity. Variants reported by the current study include missense variants, truncating nonsense variants, and an intragenic deletion. Clinical features were previously described and included developmental delay, learning difficulties, microcephaly, macrocephaly, seizures, behavioral issues (aggression, stereotypies), skeletal problems including short, tapering fingers and scoliosis, dental problems (overcrowding/delayed eruption), and variable facial features. Here, we report clinical features that have not been described previously, including specific structural brain malformations such as abnormalities of the corpus callosum and ventriculomegaly, additional psychological and dental issues along with a more recognizable facial gestalt linked to the specific domains of the TRIO gene and the effect of the variant upon the function of the encoded protein. This current study further strengthens the genotype-to-phenotype correlation that was previously established and extends the range of phenotypes to include structural brain abnormalities, additional skeletal, dental, and psychiatric issues.

A panel-agnostic strategy 'HiPPo' improves diagnostic efficiency in the UK Genome Medicine Service.

Genome sequencing is now available as a clinical test on the National Health Service (NHS) through the Genome Medicine Service (GMS). The GMS have set out an analytical strategy that predominantly filters genome data on a pre-selected gene panel(s). Whilst this approach reduces the number of variants requiring assessment by reporting laboratories, pathogenic variants outside of the gene panel applied may be missed, and candidate variants in novel genes are largely ignored. This study sought to compare a research exome analysis to an independent clinical genome analysis performed through the NHS for the same group of patients. When analysing the exome data, we applied a panel agnostic approach filtering for variants with High Pathogenic Potential (HiPPo) using ClinVar, allele frequency, and in silico prediction tools. We then compared this gene agnostic analysis to the panel-based approach as applied by the GMS to genome data. Later we restricted HiPPo variants to a panel of the Gene Curation Coalition (GenCC) morbid genes and compared the diagnostic yield with the variants filtered using the GMS strategy. 24 patients from 8 families underwent parallel research exome sequencing and GMS genome sequencing. HiPPo analysis applied to research exome data identified a similar number of variants as the gene panel-based approach applied by the GMS. GMS clinical genome analysis identified and returned 2 pathogenic variants and 3 variants of uncertain significance. HiPPo research exome analysis identified the same variants plus an additional pathogenic variant and a further 3 de novo variants of uncertain significance in novel genes, where case series and functional studies are underway. When HiPPo was restricted to GenCC disease genes (strong or definitive), the same pathogenic variants were identified yet statistically fewer variants required assessment to identify more diagnostic variants than reported by the GMS genome strategy. This gave a diagnostic rate per variant assessed of 20% for HiPPo restricted to GenCC versus 3% for the GMS panel-based approach. With plans to sequence 5 million more NHS patients, strategies are needed to optimise the full potential of genome data beyond gene panels whilst minimising the burden of variants that require clinical assessment.

Pathogenic TRIO variants associated with neurodevelopmental disorders perturb the molecular regulation of TRIO and axon pathfinding in vivo.

The RhoGEF TRIO is known to play a major role in neuronal development by controlling actin cytoskeleton remodeling, primarily through the activation of the RAC1 GTPase. Numerous de novo mutations in the TRIO gene have been identified in individuals with neurodevelopmental disorders (NDDs). We have previously established the first phenotype/genotype correlation in TRIO-associated diseases, with striking correlation between the clinical features of the individuals and the opposite modulation of RAC1 activity by TRIO variants targeting different domains. The mutations hyperactivating RAC1 are of particular interest, as they are recurrently found in patients and are associated with a severe form of NDD and macrocephaly, indicating their importance in the etiology of the disease. Yet, it remains unknown how these pathogenic TRIO variants disrupt TRIO activity at a molecular level and how they affect neurodevelopmental processes such as axon outgrowth or guidance. Here we report an additional cohort of individuals carrying a pathogenic TRIO variant that reinforces our initial phenotype/genotype correlation. More importantly, by performing conformation predictions coupled to biochemical validation, we propose a model whereby TRIO is inhibited by an intramolecular fold and NDD-associated variants relieve this inhibition, leading to RAC1 hyperactivation. Moreover, we show that in cultured primary neurons and in the zebrafish developmental model, these gain-of-function variants differentially affect axon outgrowth and branching in vitro and in vivo, as compared to loss-of-function TRIO variants. In summary, by combining clinical, molecular, cellular and in vivo data, we provide compelling new evidence for the pathogenicity of novel genetic variants targeting the TRIO gene in NDDs. We report a novel mechanism whereby the fine-tuned regulation of TRIO activity is critical for proper neuronal development and is disrupted by pathogenic mutations.

Overlapping cortical malformations in patients with pathogenic variants in GRIN1 and GRIN2B.

BACKGROUND: Malformations of cortical development (MCDs) have been reported in a subset of patients with pathogenic heterozygous variants in GRIN1 or GRIN2B, genes which encode for subunits of the N-methyl-D-aspartate receptor (NMDAR). The aim of this study was to further define the phenotypic spectrum of NMDAR-related MCDs. METHODS: We report the clinical, radiological and molecular features of 7 new patients and review data on 18 previously reported individuals with NMDAR-related MCDs. Neuropathological findings for two individuals with heterozygous variants in GRIN1 are presented. We report the clinical and neuropathological features of one additional individual with homozygous pathogenic variants in GRIN1. RESULTS: Heterozygous variants in GRIN1 and GRIN2B were associated with overlapping severe clinical and imaging features, including global developmental delay, epilepsy, diffuse dysgyria, dysmorphic basal ganglia and hippocampi. Neuropathological examination in two fetuses with heterozygous GRIN1 variants suggests that proliferation as well as radial and tangential neuronal migration are impaired. In addition, we show that neuronal migration is also impaired by homozygous GRIN1 variants in an individual with microcephaly with simplified gyral pattern. CONCLUSION: These findings expand our understanding of the clinical and imaging features of the 'NMDARopathy' spectrum and contribute to our understanding of the likely underlying pathogenic mechanisms leading to MCD in these patients.

Biallelic PRMT7 pathogenic variants are associated with a recognizable syndromic neurodevelopmental disorder with short stature, obesity, and craniofacial and digital abnormalities.

PURPOSE: Protein arginine methyltransferase 7 (PRMT7) is a member of a family of enzymes that catalyzes the methylation of arginine residues on several protein substrates. Biallelic pathogenic PRMT7 variants have previously been associated with a syndromic neurodevelopmental disorder characterized by short stature, brachydactyly, intellectual developmental disability, and seizures. To our knowledge, no comprehensive study describes the detailed clinical characteristics of this syndrome. Thus, we aim to delineate the phenotypic spectrum of PRMT7-related disorder. METHODS: We assembled a cohort of 51 affected individuals from 39 different families, gathering clinical information from 36 newly described affected individuals and reviewing data of 15 individuals from the literature. RESULTS: The main clinical characteristics of the PRMT7-related syndrome are short stature, mild to severe developmental delay/intellectual disability, hypotonia, brachydactyly, and distinct facial morphology, including bifrontal narrowing, prominent supraorbital ridges, sparse eyebrows, short nose with full/broad nasal tip, thin upper lip, full and everted lower lip, and a prominent or squared-off jaw. Additional variable findings include seizures, obesity, nonspecific magnetic resonance imaging abnormalities, eye abnormalities (i.e., strabismus or nystagmus), and hearing loss. CONCLUSION: This study further delineates and expands the molecular, phenotypic spectrum and natural history of PRMT7-related syndrome characterized by a neurodevelopmental disorder with skeletal, growth, and endocrine abnormalities.

Targeting de novo loss-of-function variants in constrained disease genes improves diagnostic rates in the 100,000 Genomes Project.

BACKGROUND: Genome sequencing was first offered clinically in the UK through the 100,000 Genomes Project (100KGP). Analysis was restricted to predefined gene panels associated with the patient's phenotype. However, panels rely on clearly characterised phenotypes and risk missing diagnoses outside of the panel(s) applied. We propose a complementary method to rapidly identify pathogenic variants, including those missed by 100KGP methods. METHODS: The Loss-of-function Observed/Expected Upper-bound Fraction (LOEUF) score quantifies gene constraint, with low scores correlated with haploinsufficiency. We applied DeNovoLOEUF, a filtering strategy to sequencing data from 13,949 rare disease trios in the 100KGP, by filtering for rare, de novo, loss-of-function variants in disease genes with a LOEUF score < 0.2. We compared our findings with the corresponding patient's diagnostic reports. RESULTS: 324/332 (98%) of the variants identified using DeNovoLOEUF were diagnostic or partially diagnostic (whereby the variant was responsible for some of the phenotype). We identified 39 diagnoses that were "missed" by 100KGP standard analyses, which are now being returned to patients. CONCLUSION: We have demonstrated a highly specific and rapid method with a 98% positive predictive value that has good concordance with standard analysis, low false-positive rate, and can identify additional diagnoses. Globally, as more patients are being offered genome sequencing, we anticipate that DeNovoLOEUF will rapidly identify new diagnoses and facilitate iterative analyses when new disease genes are discovered.

Blood gene expression predicts intensive care unit admission in hospitalised patients with COVID-19.

Background: The COVID-19 pandemic has created pressure on healthcare systems worldwide. Tools that can stratify individuals according to prognosis could allow for more efficient allocation of healthcare resources and thus improved patient outcomes. It is currently unclear if blood gene expression signatures derived from patients at the point of admission to hospital could provide useful prognostic information. Methods: Gene expression of whole blood obtained at the point of admission from a cohort of 78 patients hospitalised with COVID-19 during the first wave was measured by high resolution RNA sequencing. Gene signatures predictive of admission to Intensive Care Unit were identified and tested using machine learning and topological data analysis, TopMD. Results: The best gene expression signature predictive of ICU admission was defined using topological data analysis with an accuracy: 0.72 and ROC AUC: 0.76. The gene signature was primarily based on differentially activated pathways controlling epidermal growth factor receptor (EGFR) presentation, Peroxisome proliferator-activated receptor alpha (PPAR-α) signalling and Transforming growth factor beta (TGF-ß) signalling. Conclusions: Gene expression signatures from blood taken at the point of admission to hospital predicted ICU admission of treatment naïve patients with COVID-19.

Clinical, splicing, and functional analysis to classify BRCA2 exon 3 variants: Application of a points-based ACMG/AMP approach.

Skipping of BRCA2 exon 3 (∆E3) is a naturally occurring splicing event, complicating clinical classification of variants that may alter ∆E3 expression. This study used multiple evidence types to assess pathogenicity of 85 variants in/near BRCA2 exon 3. Bioinformatically predicted spliceogenic variants underwent mRNA splicing analysis using minigenes and/or patient samples. ∆E3 was measured using quantitative analysis. A mouse embryonic stem cell (mESC) based assay was used to determine the impact of 18 variants on mRNA splicing and protein function. For each variant, population frequency, bioinformatic predictions, clinical data, and existing mRNA splicing and functional results were collated. Variant class was assigned using a gene-specific adaptation of ACMG/AMP guidelines, following a recently proposed points-based system. mRNA and mESC analysis combined identified six variants with transcript and/or functional profiles interpreted as loss of function. Cryptic splice site use for acceptor site variants generated a transcript encoding a shorter protein that retains activity. Overall, 69/85 (81%) variants were classified using the points-based approach. Our analysis shows the value of applying gene-specific ACMG/AMP guidelines using a points-based approach and highlights the consideration of cryptic splice site usage to appropriately assign PVS1 code strength.

Recommendations for clinical interpretation of variants found in non-coding regions of the genome.

BACKGROUND: The majority of clinical genetic testing focuses almost exclusively on regions of the genome that directly encode proteins. The important role of variants in non-coding regions in penetrant disease is, however, increasingly being demonstrated, and the use of whole genome sequencing in clinical diagnostic settings is rising across a large range of genetic disorders. Despite this, there is no existing guidance on how current guidelines designed primarily for variants in protein-coding regions should be adapted for variants identified in other genomic contexts. METHODS: We convened a panel of nine clinical and research scientists with wide-ranging expertise in clinical variant interpretation, with specific experience in variants within non-coding regions. This panel discussed and refined an initial draft of the guidelines which were then extensively tested and reviewed by external groups. RESULTS: We discuss considerations specifically for variants in non-coding regions of the genome. We outline how to define candidate regulatory elements, highlight examples of mechanisms through which non-coding region variants can lead to penetrant monogenic disease, and outline how existing guidelines can be adapted for the interpretation of these variants. CONCLUSIONS: These recommendations aim to increase the number and range of non-coding region variants that can be clinically interpreted, which, together with a compatible phenotype, can lead to new diagnoses and catalyse the discovery of novel disease mechanisms.

A systematic analysis of splicing variants identifies new diagnoses in the 100,000 Genomes Project.

BACKGROUND: Genomic variants which disrupt splicing are a major cause of rare genetic diseases. However, variants which lie outside of the canonical splice sites are difficult to interpret clinically. Improving the clinical interpretation of non-canonical splicing variants offers a major opportunity to uplift diagnostic yields from whole genome sequencing data. METHODS: Here, we examine the landscape of splicing variants in whole-genome sequencing data from 38,688 individuals in the 100,000 Genomes Project and assess the contribution of non-canonical splicing variants to rare genetic diseases. We use a variant-level constraint metric (the mutability-adjusted proportion of singletons) to identify constrained functional variant classes near exon-intron junctions and at putative splicing branchpoints. To identify new diagnoses for individuals with unsolved rare diseases in the 100,000 Genomes Project, we identified individuals with de novo single-nucleotide variants near exon-intron boundaries and at putative splicing branchpoints in known disease genes. We identified candidate diagnostic variants through manual phenotype matching and confirmed new molecular diagnoses through clinical variant interpretation and functional RNA studies. RESULTS: We show that near-splice positions and splicing branchpoints are highly constrained by purifying selection and harbour potentially damaging non-coding variants which are amenable to systematic analysis in sequencing data. From 258 de novo splicing variants in known rare disease genes, we identify 35 new likely diagnoses in probands with an unsolved rare disease. To date, we have confirmed a new diagnosis for six individuals, including four in whom RNA studies were performed. CONCLUSIONS: Overall, we demonstrate the clinical value of examining non-canonical splicing variants in individuals with unsolved rare diseases.