The fast Viterbi algorithm caching Profile Hidden Markov Models on graphic processing units

Profile Hidden Markov Models are used as a popular tool in bioinformatics research and a regular task is to compare a set of protein sequences with a database of models according to sequences' score on these models. However, it suffers from long runtimes on PC platforms, and the runtimes are increasing rapidly due to the rapid growth in size of both sequences and models. In this paper, we present a Viterbi algorithm running on graphic processing units to score sequences, a method padding HMMs and a memory hierarchy are also introduced, these strategies can promote running efficiency in parallel and reduce impact of the bottleneck from buses. Experimental results show the runtimes are saved by the method dramatically.