Machine learning in genome‐wide association studies

Abstract

Recently, genome-wide association studies have substantially expanded our knowledge about genetic variants that influence the susceptibility to complex diseases. Although standard statistical tests for each single-nucleotide polymorphism (SNP) separately are able to capture main genetic effects, different approaches are necessary to identify SNPs that influence disease risk jointly or in complex interactions. Experimental and simulated genome-wide SNP data provided by the Genetic Analysis Workshop 16 afforded an opportunity to analyze the applicability and benefit of several machine learning methods. Penalized regression, ensemble methods, and network analyses resulted in several new findings while known and simulated genetic risk variants were also identified. In conclusion, machine learning approaches are promising complements to standard single-and multi-SNP analysis methods for understanding the overall genetic architecture of complex human diseases. However, because they are not optimized for genome-wide SNP data, improved implementations and new variable selection procedures are required.

Overview of random forest methodology and practical guidance with emphasis on computational biology and bioinformatics

“…This good performance is documented in several independent comparison studies implementing different simulation settings [26,54,73,17]. In these studies, however, standard VIMs (either Gini or permutation) are used to rank the SNPs.…”

Polygenic scores via penalized regression on summary statistics

“…In principle, various machine learning methods or Bayesian methods can be applied in the construction of PGS, as they have been applied in the estimation of breeding values in animal studies (Meuwissen et al, 2001;Abraham et al, 2013;Szymczak et al, 2009;Habier et al, 2011;Pirinen et al, 2013;Erbe et al, 2012;Ogutu et al, 2012;Zhou et al, 2013). These methods do not require the assumption of SNP independence or near independence, and have been shown to perform better than simple PGS in simulation settings.…”

Detecting epistasis in the presence of linkage disequilibrium: A focused comparison

“…The literature contains many reviews of SNP-SNP interaction models [17] [1], [18], [19], [2], [20], [21], [22]. These reviews pinpoint the strengths and weaknesses of existing methods with respect to 1) ability to detect interactions when no main effects are present 2) computational efficiency 3) the quality of the detected interactions 4) the ability to deal with higher order interactions (i.e.…”