Accelerating Genomic Analyses with Parallel Sliding Windows

Abstract

In recent years biology has become an information science, where an avalanche of newly sequenced genomic data has overwhelmed our existing analysis and mining tools. This paper addresses this challenge by developing a systematic way of speeding up a broad class of bioinformatics algorithms using commodity graphics pro- cessing hardware. Using the example problem of analyzing DNA structural variations, we demonstrate how such computations can be significantly accelerated in various parallel architectures, yield- ing over two orders of magnitude speedups at low cost and with rel- atively modest programming effort. Our implementation of a slid- ing window -based technique on the GPU and Cell architectures seems promising in its generality and extensibility to other prob- lems and domains.