Phenome-wide identification of therapeutic genetic targets, leveraging knowledge graphs, graph neural networks, and UK Biobank data.

The ongoing expansion of human genomic datasets propels therapeutic target identification; however, extracting gene-disease associations from gene annotations remains challenging. Here, we introduce Mantis-ML 2.0, a framework integrating AstraZeneca's Biological Insights Knowledge Graph and numerous tabular datasets, to assess gene-disease probabilities throughout the phenome. We use graph neural networks, capturing the graph's holistic structure, and train them on hundreds of balanced datasets via a robust semi-supervised learning framework to provide gene-disease probabilities across the human exome. Mantis-ML 2.0 incorporates natural language processing to automate disease-relevant feature selection for thousands of diseases. The enhanced models demonstrate a 6.9% average classification power boost, achieving a median receiver operating characteristic (ROC) area under curve (AUC) score of 0.90 across 5220 diseases from Human Phenotype Ontology, OpenTargets, and Genomics England. Notably, Mantis-ML 2.0 prioritizes associations from an independent UK Biobank phenome-wide association study (PheWAS), providing a stronger form of triaging and mitigating against underpowered PheWAS associations. Results are exposed through an interactive web resource.

Effects of protein-coding variants on blood metabolite measurements and clinical biomarkers in the UK Biobank.

Genome-wide association studies (GWASs) have established the contribution of common and low-frequency variants to metabolic blood measurements in the UK Biobank (UKB). To complement existing GWAS findings, we assessed the contribution of rare protein-coding variants in relation to 355 metabolic blood measurements-including 325 predominantly lipid-related nuclear magnetic resonance (NMR)-derived blood metabolite measurements (Nightingale Health Plc) and 30 clinical blood biomarkers-using 412,393 exome sequences from four genetically diverse ancestries in the UKB. Gene-level collapsing analyses were conducted to evaluate a diverse range of rare-variant architectures for the metabolic blood measurements. Altogether, we identified significant associations (p < 1 × 10-8) for 205 distinct genes that involved 1,968 significant relationships for the Nightingale blood metabolite measurements and 331 for the clinical blood biomarkers. These include associations for rare non-synonymous variants in PLIN1 and CREB3L3 with lipid metabolite measurements and SYT7 with creatinine, among others, which may not only provide insights into novel biology but also deepen our understanding of established disease mechanisms. Of the study-wide significant clinical biomarker associations, 40% were not previously detected on analyzing coding variants in a GWAS in the same cohort, reinforcing the importance of studying rare variation to fully understand the genetic architecture of metabolic blood measurements.

DrugnomeAI is an ensemble machine-learning framework for predicting druggability of candidate drug targets.

The druggability of targets is a crucial consideration in drug target selection. Here, we adopt a stochastic semi-supervised ML framework to develop DrugnomeAI, which estimates the druggability likelihood for every protein-coding gene in the human exome. DrugnomeAI integrates gene-level properties from 15 sources resulting in 324 features. The tool generates exome-wide predictions based on labelled sets of known drug targets (median AUC: 0.97), highlighting features from protein-protein interaction networks as top predictors. DrugnomeAI provides generic as well as specialised models stratified by disease type or drug therapeutic modality. The top-ranking DrugnomeAI genes were significantly enriched for genes previously selected for clinical development programs (p value < 1 × 10-308) and for genes achieving genome-wide significance in phenome-wide association studies of 450 K UK Biobank exomes for binary (p value = 1.7 × 10-5) and quantitative traits (p value = 1.6 × 10-7). We accompany our method with a web application ( http://drugnomeai.public.cgr.astrazeneca.com ) to visualise the druggability predictions and the key features that define gene druggability, per disease type and modality.

Analysis of 983 civilian blast and ballistic casualties and the generation of a template of injury burden: An observational study.

Background: Terrorism and armed conflict cause blast and ballistic casualties that are unusual in civilian practice. The immediate surgical response to mass casualty events, with civilians injured by these mechanisms, has not been systematically characterised. Standardising an approach to reacting to these events is challenging but is essential to optimise preparation for them. We aimed to quantify and assesses the surgical response to blast and ballistic injuries managed in a world-class trauma unit paradigm. Methods: This was an observational study conducted at the UK-led military Medical Treatment Facility, Camp Bastion, Afghanistan from original theatre log-book entries between Nov 5, 2009, and Sept 21, 2014; a total of 10,891 consecutive surgical cases prospectively gathered by surgical teams were catalogued. Patients with combatant status/wearing body-armour to various degrees including interpreters were excluded from the study. Civilian casualties that underwent primary trauma surgery for blast and ballistic injuries were included (n=983). Surgical activity was analysed as a rate per 100 casualties, and patients were grouped according to adult vs. paediatric and ballistic vs. blast injury mechanisms to aid comparison. Findings: The three most common surgical procedures for civilian blast injuries were debridement, amputation, and laparotomy. For civilian ballistic injuries, these were debridement, laparotomy and vascular procedures. Blast injuries generated more amputations in both adults and children compared to ballistic injuries. Blast injuries generated more removal of fragmentation material compared to ballistics injuries amongst adult casualties. Ballistic injuries lead to more chest drain insertions in adults. As a rate per 100 casualties, adults injured by blast underwent significantly more debridement (63·5); temporary skeletal stabilisation (13·2) and vascular procedures (12·8) compared to children (43·4, z=4·026, p=0·00007; 5·7, z=2·230, p=0·022; 4·9, z=2·468, p=0·014). Adults injured by ballistics underwent significantly more debridement (63·4); chest drain (12·3) and temporary skeletal fixation procedures (11·4) compared to children (50·0, z=2·058, p=0.040, p<0·05; 2·9, z=2·283, p=0.0230; 2·9, z=2·131, p=0.034 respectively). By comparison, children injured by ballistics underwent significantly more removal of fragmentation and ballistic materials (20·6) when compared to adults (7·7, z=-3·234; p=0.001). Interpretation: This is the first evidence-based, template of the immediate response required to manage civilians injured by blast and ballistic mechanisms. The template presented can be applied to similar conflict zones and to prepare for terror attacks on urban populations. Funding: The work was supported in part by a grant to LM from School of Medicine, University of St Andrews.

Gene-SCOUT: identifying genes with similar continuous trait fingerprints from phenome-wide association analyses.

Large-scale phenome-wide association studies performed using densely-phenotyped cohorts such as the UK Biobank (UKB), reveal many statistically robust gene-phenotype relationships for both clinical and continuous traits. Here, we present Gene-SCOUT, a tool used to identify genes with similar continuous trait fingerprints to a gene of interest. A fingerprint reflects the continuous traits identified to be statistically associated with a gene of interest based on multiple underlying rare variant genetic architectures. Similarities between genes are evaluated by the cosine similarity measure, to capture concordant effect directionality, elucidating clusters of genes in a high dimensional space. The underlying gene-biomarker population-scale association statistics were obtained from a gene-level rare variant collapsing analysis performed on over 1500 continuous traits using 394 692 UKB participant exomes, with additional metabolomic trait associations provided through Nightingale Health's recent study of 121 394 of these participants. We demonstrate that gene similarity estimates from Gene-SCOUT provide stronger enrichments for clinical traits compared to existing methods. Furthermore, we provide a fully interactive web-resource (http://genescout.public.cgr.astrazeneca.com) to explore the pre-calculated exome-wide similarities. This resource enables a user to examine the biological relevance of the most similar genes for Gene Ontology (GO) enrichment and UKB clinical trait enrichment statistics, as well as a detailed breakdown of the traits underpinning a given fingerprint.

Prioritizing non-coding regions based on human genomic constraint and sequence context with deep learning.

Elucidating functionality in non-coding regions is a key challenge in human genomics. It has been shown that intolerance to variation of coding and proximal non-coding sequence is a strong predictor of human disease relevance. Here, we integrate intolerance to variation, functional genomic annotations and primary genomic sequence to build JARVIS: a comprehensive deep learning model to prioritize non-coding regions, outperforming other human lineage-specific scores. Despite being agnostic to evolutionary conservation, JARVIS performs comparably or outperforms conservation-based scores in classifying pathogenic single-nucleotide and structural variants. In constructing JARVIS, we introduce the genome-wide residual variation intolerance score (gwRVIS), applying a sliding-window approach to whole genome sequencing data from 62,784 individuals. gwRVIS distinguishes Mendelian disease genes from more tolerant CCDS regions and highlights ultra-conserved non-coding elements as the most intolerant regions in the human genome. Both JARVIS and gwRVIS capture previously inaccessible human-lineage constraint information and will enhance our understanding of the non-coding genome.