A Novel Approach Based on Relevance Learning Vector Quantization Applied to the Inference of High-Order SNPs Interactions

Abstract

Single Nucleotide Polymorphism (SNP) is the most common variation present in the genome. Correlated SNPs interactions, called SNPs interaction, have been associated with different phenotype expressions, such as increased risk of some complex diseases. Computationally identifying SNPs interactions involving more than two SNPs, which are called high-order interactions, has proven to be a challenging problem. In this paper, we propose a new approach, called Incremental Generalized Relevance Learning Vector Quantization for SNP inference (iGRLVQ-SNPi), that can efficiently infer SNPs involved in high-order interactions with high accuracy. In our experiments, iGRLVQ-SNPi obtained very good results and it achieves better results than the other methods compared. Our innovation to the learning process of LVQ is a promising approach for inference of SNPs interactions in high-order.