Machine learning prioritized 508 genes likely causal for HCM, identifying 113 drug-interacting genes including IRS1 and ADH1A as potential repurposing targets.
A machine learning approach identified 508 genes potentially causal for hypertrophic cardiomyopathy, highlighting novel genetic targets and drug repurposing opportunities.
Absolute Event Rate: 0% vs 0%
Abstract Background Hypertrophic cardiomyopathy (HCM) is a complex, heterogeneous genetic disorder characterized by left ventricular hypertrophy and impaired cardiac function, with limited targeted treatment options. Genetic studies can highlight new druggable targets and drug repurposing opportunities, for a more personalised approach to HCM treatment. Purpose This study aims to identify and prioritise novel genes associated with HCM, assess their involvement in cardiomyopathy-related pathways, and explore potential drug targets and drug repurposing opportunities. Methods This study uses Mantis-ML2 machine learning software to systematically prioritise HCM-related genes. Mantis-ML2 takes as input gene probability scores generated by a graph convolutional network (GCN) that integrates knowledge graphs, natural language processing, and biological features (e.g. gene expression, tissue specificity, and protein interactions) to prioritise the most likely HCM causal genes (Figure 1). The model was augmented to include a bespoke feature from Exomiser, annotating the most likely HCM genes based on clinical and animal model data. From the output gene prioritisation, a confidence threshold of ≥0.75 was used to select the most likely causal HCM genes. These prioritised genes were cross-checked against drug interactions in Open Targets. Gene enrichment analysis tools (gProfiler, WebGestalt and Enrichr) and the International Mouse Phenotyping Consortium (IMPC) database were used to identify significant pathways and mouse model phenotypes. The Side Effect Resource (SIDER) was used to evaluate the risks and feasibility of repurposing identified drugs for HCM treatment. Results Mantis-ML2’s GCN had an AUC (area under the curve) of 0.87, prioritising 508 genes with a probability ≥0.75 of being causal for HCM, including 105 already linked to HCM in the GWAS catalog. Out of the 508 genes, Open Targets identified 113 drug-interacting genes. Among these, 70 genes interacted with cardiovascular disease (CVD)-associated drugs, and 43 were linked to non-CVD drugs. Gene enrichment analysis linked NDUFC2 and NDUFB6 to the oxidative phosphorylation and diabetic cardiomyopathy pathways (adjusted p-value 6.55E-32 and 4.49E-35, respectively; Mantis-ML2 probability of 0.99). It also highlighted FGF9, which is linked to the calcium signaling pathway (adjusted p-value 9.88E-5) and to enlarged heart mouse model phenotypes (adjusted p-value 8.05E-9). SIDER analysis identified potentially repurposable gene-drug targets, including IRS1 targeting Aganirsen, linked to cardiovascular risks, and ADH1A modulator Fomepizole, which may influence oxidative stress in HCM. Conclusion This study prioritises HCM genes using advanced machine learning and multi-layered evidence, highlighting their pharmacological interactions and supporting a genetics-driven approach to drug repurposing. The findings of IRS1, ADH1A, NDUFC2 and NDUFB6 bring new insights into the genetics behind HCM.
Apti et al. (Sat,) reported a other. Machine learning prioritized 508 genes likely causal for HCM, identifying 113 drug-interacting genes including IRS1 and ADH1A as potential repurposing targets.