Inferring Gene Regulatory Networks Using Hybrid Parallel Computing

Gene regulatory networks (GRN) inference is an important bioinformatics problem, with many applications in system biology, in which the gene interactions need to be deduced from gene expression data, such as microarray data or RNA-Seq. Depending on the algorithm, the inference process may take a large amount of time due to the size of the networks and the complexity of the algorithm. The inference algorithm described in this work is based on the seed growing paradigm and has two steps: the seed growing and the inference step. We developed two parallel versions of the inference algorithm with the following approaches: Cluster of CPUs and a hybrid Multi CPU/GPUs. In tests performed in three databases of genes with different samples, these versions present good speedups, scalability, and robustness when compared with the sequential algorithm. Namely, for the seed growing step we achieved speedup of 12.7 with 16 nodes on Cluster of CPUs and speedup of 12.3 on hybrid Multi CPU/GPUs approach using 8 CPU/GPUs. For the inference step, we achieved speedups up to 10 on Cluster of CPUs.