ABSTRACT
Motivation In computing pairwise alignments of biological sequences, software implementations employ a variety of heuristics that decrease the computational effort involved in computing potential alignments. A key element in achieving high processing throughput is to identify and prioritize potential alignments where high-scoring mappings can be expected. These tasks involve listprocessing operations that can be efficiently performed on GPU hardware.
Results We implemented a read aligner called A21 that exploits GPU-based parallel sort and reduction techniques to restrict the number of locations where potential alignments may be found. When compared with other high-throughput aligners, this approach finds more high-scoring mappings without sacrificing speed or accuracy. A21 running on a single GPU is about 10 times faster than comparable CPU-based tools; it is also faster and more sensitive in comparison with other recent GPU-based aligners.
Availability The A21 software is open source and available at https://github.com/RWilton/A21.
Contact rwilton{at}pha.jhu.edu
Footnotes
↵* To whom correspondence should be addressed.