gpuZoo: Cost-effective estimation of gene regulatory networks using the Graphics Processing Unit

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Gene regulatory network inference models genome-scale regulatory processes crucial for development, disease, and response to perturbations.
• Existing tools like PANDA, SPIDER, PUMA, and LIONESS model transcriptional and post-transcriptional regulation but face computational complexity limitations.
• High computational costs and run times restrict the application scope of current gene regulatory network inference methods.

Purpose of the Study:

• To enhance the cost and time performance of established gene regulatory network inference algorithms.
• To develop a GPU-accelerated implementation for significantly improving computational efficiency.

Main Methods:

• Development of gpuZoo, a software package implementing GPU-accelerated calculations for gene regulatory network inference.
• Integration of gpuZoo with existing tools, enabling faster processing of data matrices for network structure optimization.
• Implementation in both MATLAB (netZooM) and Python (netZooPy) for broad accessibility.

Main Results:

• The gpuZoo implementation achieves up to 61 times faster runtimes compared to multi-core CPU implementations.
• Computational costs are reduced by up to 28 times with the GPU-accelerated approach.
• Demonstrates dramatic improvements in performance for modeling transcriptional and post-transcriptional gene regulation.

Conclusions:

• gpuZoo significantly overcomes the computational limitations of existing gene regulatory network inference methods.
• The GPU-accelerated approach makes large-scale regulatory process modeling more accessible and cost-effective.
• Accelerated inference enables broader applications in studying complex biological systems.