Searching Positive Selection using GPU and Benchmarking GPU Based Short Read Aligners

Abstract

This study and thesis is divided into two parts. The first part of this thesis pertains to development of an algorithm for searching positive selection(specifically by implementing Branch Site Model) using GPU and ascertain the feasibility of such implementation. This include changes in the underlying algorithm of preexisting tools to accommodate GPU and HPC acceleration paradigms. Along with implementation, this study also identified potential drawbacks of such implementation, and alternative strategies for possible program. The second part of this thesis deals with performance profiling of short read aligners (using Nvidia CUDA framework) for testing scalability in GPGPU environment. We studied the impact of GPU based aligners on NGS analysis pipeline and included a comparison with CPU based counterparts. The main aim of this study was to identify the possible gains by using GPU in NGS analysis within a similar price bracket and to study the implementation of such aligners in GPU. The performance was measured by running alignments on simulated Illumina reads on human genome.