GPU-accelerated bidirected De Bruijn graph construction for genome assembly

Abstract

De Bruijn graph construction is a basic component in de novo genome assembly for short reads generated from the second-generation sequencing machines. As this component processes a large amount of data and performs intensive computation, we propose to use the GPU (Graphics Processing Unit) for acceleration. Specifically, we propose a staged algorithm to utilize the GPU for computation over large data sets that do not fit into the GPU memory. We also pipeline the I/O, GPU, and CPU processing to further improve the overall performance. Our preliminary results show that our GPU-accelerated graph construction on an NVIDIA S1070 server achieves a speedup of around two times over previous performance results on a 1024-node IBM Blue Gene/L. © 2013 Springer-Verlag.