GPU-accelerated Burrow-Wheeler Transform for genomic data

Abstract

As the core of FM-index and compressed suffix array, the Burrows-Wheeler Transform (BWT) plays a key role in indexing genomic sequence data for pattern search. It is more space-efficient than suffix tree and suffix array and also has good searching performance. However, computing a BWT from a sequence in efficient space and time is challenging. In this article, we present a fast algorithm for computing BWT which leverages the computing power of graphics processing unit (GPU). It is based on a blockwise suffix sorting method which builds BWT block-by-block and uses a GPU to accelerate the suffix sorting procedures. Our GPU-accelerated algorithm only needs small memory space and experiments show that it is about 16 times faster than the original CPU implementation for computing BWT of human whole genome.