High-Order Epistasis Detection in High Performance Computing Systems

Resumen

In recent years, Genome-Wide Association Studies (GWAS) have become more and more popular with the intent of finding a genetic explanation for the presence or absence of particular diseases in human studies. There is consensus about the presence of genetic interactions during the expression of complex diseases, a phenomenon called epistasis. This thesis focuses on the study of this phenomenon, employingHigh- Performance Computing (HPC) for this purpose and from a statistical definition of the problem: the deviation of the expression of a phenotype from the addition of the individual contributions of genetic variants. For this purpose, we first developedMPI3SNP, a programthat identifies interactions of three variants froman input dataset. MPI3SNP implements an exhaustive search of epistasis using an association test based on the Mutual Information and exploits the resources of clusters of CPUs or GPUs to speed up the search. Then, we evaluated the state-of-the-art methods with the help of MPI3SNP in a study that compares the performance of twenty-seven tools. The most important conclusion of this study is the inability of non-exhaustive approaches to locate epistasis in the absence of marginal effects (small association effects of individual variants that partake in an epistasis interaction). For this reason, this thesis continued focusing on the optimization of the exhaustive search. First, we improved the efficiency of the association test through a vector implementation of this procedure. Then, we developed a distributed algorithm capable of locating epistasis interactions of any order. These two milestones were achieved in Fiuncho, a program that incorporates all the research carried out, obtaining the best performance in CPU clusters out of all the alternatives of the state-of-the-art. In addition, we also developed a library to simulate particular scenarios with epistasis called Toxo. This library allows for the simulation of epistasis that follows existing interaction models for high-order interactions.