High performance computing architectures analysis for gene networks inference

Modeling and inference of biological systems are an important field in computer science, presenting strong interdisciplinary aspects. In this context, the inference of gene regulatory networks and the analysis of their dynamics generated by their transition functions are important issues that demand substantial computational power. Because the algorithms that return the optimal solution have an exponential time cost, such algorithms only work for gene networks with only dozens of genes. However realistic gene networks present hundreds to thousands of genes, with some genes being hubs, i.e., their number of predictor genes are usually much higher than average. Therefore there is a need to develop ways to speed up the gene networks inference. This paper presents a benchmark involving GPUs and FPGAs to infer gene networks, analysing processing time, hardware cost acquisition, energy consumption and programming complexity. Overall Titan XP GPU achieved the best performance, but with a large cost regarding acquisition price when compared to R9 Nano GPU and DE1-SOC FPGA. In its turn, R9 Nano GPU presented the best cost-benefit regarding performance, acquisition price, energy consumption, and programming complexity, although DE1-SOC FPGA presented much smaller energy consumption.