Genome-wide association studies of ischemic stroke based on interpretable machine learning

Area of Science:

• Genetics
• Computational Biology
• Neurology

Background:

• Genetic underpinnings of ischemic stroke (IS) are not fully understood despite numerous identified loci.
• Existing genome-wide association studies (GWAS) require advanced analytical methods for comprehensive genetic discovery.

Purpose of the Study:

• To identify novel genetic associations with ischemic stroke (IS).
• To explore the utility of machine learning algorithms in analyzing large-scale genotypic data for complex diseases.
• To establish a consensus approach for integrating results from diverse analytical techniques.

Main Methods:

• Genome-wide association studies (GWAS) employing classical statistical testing.
• Application of machine learning algorithms: logistic regression, gradient boosting on decision trees, and TabNet.
• Utilized Pareto-Optimal solutions to consolidate findings from different analytical methods.
• Analyzed genotypic data from 5,581 individuals of European ancestry (883,749 SNPs) from the Database of Genotypes and Phenotypes.

Main Results:

• Identified 131 candidate genes associated with the onset of ischemic stroke (IS).
• Confirmed previously known IS-associated genes (UBQLN1, TRPS1, MUSK) in model animals.
• Discovered a novel association between ACOT11, involved in fatty acid metabolism, and IS.
• Highlighted GPR26 as a potential therapeutic target for stroke prevention based on its G-coupled protein receptor function.

Conclusions:

• Advanced GWAS and machine learning approaches effectively identify genetic factors in ischemic stroke (IS).
• The study expands the known genetic landscape of IS and proposes novel therapeutic avenues.
• The integrated analytical framework can be applied to GWAS datasets for other complex diseases.