Deep Learning and Genome-Wide Association Studies for the Classification of Type 2 Diabetes

Genome-wide association studies (GWAS) have promised to significantly enhance our understanding of genetic based determinants of common complex diseases. A strong body of evidence suggested that genetic factors contribute significantly to the predisposition of Type 2 Diabetes (T2D). However, many studies have shown that single-locus analysis has demonstrated little effect in understanding the genetic architecture of complex human diseases, as is the case of GWAS. Traditional machine learning models, such as random forest and support vector machine have been widely used with genome-wide data as an alternative approach. However, there are still several challenges in modelling high-dimensional GWAS data. This paper addresses these issues using a deep learning framework to model the cumulative effects of Single Nucleotide Polymorphisms (SNP) for the classification of Type 2 Diabetes in the context of genome-wide data. The findings show that using 6609 SNPs it is possible to obtain (AUC=96.53%, Sens=93.91%, Spec=90.83%, Logloss= 32.33%, Gini=93.06%, MSE=9.50%). Using a deep learning approach, it is possible to capture the latent representation of genetic variants and the important interactions between them. Our approach holds great promise and warrants further study.