A high performance reconfigurable core for motif searching using profile HMM

A common task in bioinformatics is the comparison of biological sequences to probabilistic models in order to evaluate their similarity. The HMMer software is a popular tool for aligning biological sequences to HMM models that represent biologic motifs. The results of these comparisons can be used to annotate domains of an unknown sequence or locate new members of a modelled family. But as the size of sequence and model repositories increases the time required to perform searches between entire databases becomes impractical. The HMMer engine is an example of this where extended searches can take up to days even with algorithmic improvements and software enhancements. This work describes the acceleration of the Viterbi decoding process by means of parallelizing the algorithm and mapping it to a systolic array. The concurrency of the arraypsilas processing elements is realized by implementing the engine on off-the-self FPGA hardware. Our implementation offers a 247times speedup over a software implementation of the HMMer tool on a standard desktop computer