Gene regulatory networks inference using a multi-GPU exhaustive search algorithm
BMC Bioinformatics
|February 26, 2014
Summary
We developed a low-cost, parallel exhaustive search algorithm for gene regulatory network (GRN) inference using GPUs. This approach significantly accelerates the identification of gene interactions from expression data.
Area of Science:
- Bioinformatics
- Computational Biology
- Genomics
Background:
- Gene regulatory network (GRN) inference aims to identify gene interactions from expression data.
- Exhaustive search is an accurate but computationally infeasible method for feature selection in GRN inference.
- Current methods struggle with the high computational cost of exhaustive search.
Purpose of the Study:
- To develop a cost-effective, parallelized exhaustive search algorithm for GRN inference.
- To leverage GPU architectures for accelerating complex bioinformatics problems.
- To create a viable solution for deducing gene interactions from microarray data.
Main Methods:
- Implemented a parallel algorithm utilizing multiple Graphics Processing Units (GPUs) and CUDA™.
- Applied the algorithm to gene regulatory network inference.
- Evaluated performance using single and multiple GPU configurations.
Main Results:
- Achieved significant speedups, on the order of hundreds, for GRN inference.
- Demonstrated near-linear scalability of speedup with an increasing number of GPUs.
- Validated the efficiency of the GPU-based parallel exhaustive search.
Conclusions:
- GPU parallelization of exhaustive search is a feasible and effective approach for GRN inference.
- The developed method offers a viable cost-benefit for complex combinatorial problems.
- This technique can be adapted for other bioinformatics and computational problems requiring exhaustive search.
