Section: New Results
Participants : Mathieu Giraud, Tuan Tu Tran, Jean-Stéphane Varré.
Parallelisation of PWM algorithms
We studied some parallelisations on different Position-Weight-Matrices (PWMs) algorithms. The algorithms connecting the score threshold and the Pvalue are NP-complete and rely on the enumeration on a large set of scores or words  . While the basic algorithm for finding occurrences of a PWM uncovered the best speedup, we proposed a new algorithm for the score/Pvalue computation that is more suitable for parallelization. We realized a prototype with the CUDA libraries, and report for the different problems speedups of 21× and 77× on the GPU Nvidia GTX 280  .
GPU Parallelization of ADP
Algebraic Dynamic Programming (ADP) is a framework developed at U. Bielefeld (Germany) to encode a broad range of optimization problems, including common bioinformatics problems like RNA folding or pairwise sequence alignment. During the stay of P. Steffen in Lille, we proposed a first generic parallelization of this framework. Depending on the application, we report speedups ranging from 6.1× to 25.8× on a Nvidia GTX 280  .
We started a project which intends to provide facilities for bioinformaticians to use parallel algorithms developed for the GPUs. The idea is to provide a repository for algorithms, notably written in OpenCL, and easy access through interfaces in the BioJava, BioPerl and Biopython frameworks. The project has been presented at the annual conference of bioinformatics open-source software (BOSC)  .