Genetic Variants Detection Based on Weighted Sparse Group Lasso

Abstract

Identification of associated genetic variants is a critical step both in crop breeding and in understanding complex diseases etiology and pathogenesis. Most existing methods cannot exactly identify the variants that only appear in a few target genes. In this paper, we propose a weighted sparse group lasso (WSGL) method to select common and rare variants in groups. Considering the biological assumption that complex diseases or traits are influenced by a few SNPs in a small number of genes, our method uses sparse group lasso to simultaneously select associated groups and SNPs within. To increase the chance of picking out rare variants, biological prior information is introduced into our model by re-weighting lasso regularization with weights calculated from input data. Experimental results on both simulation data and real data demonstrate the superiority of WSGL over other competitive approaches.