Inference of dynamic spatial GRN models with multi-GPU evolutionary computation

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Reverse engineering gene regulatory network (GRN) models with specific spatial dynamics is a complex inverse problem.
• Current heuristic machine learning algorithms are computationally intensive, requiring millions of model simulations and high computational resources.
• Existing methods lack GPU acceleration for inferring mechanistic GRNs from spatial phenotypes.

Purpose of the Study:

• To develop an efficient methodology for parallelizing evolutionary algorithms using GPU computing for mechanistic GRN inference.
• To enable the inference of GRNs capable of recapitulating specific gene expression patterns in multicellular systems.
• To accelerate the process of extracting biological knowledge and designing synthetic GRNs.

Main Methods:

• Parallelization of evolutionary algorithms using multi-CPU threads and asynchronous GPU kernel launches.
• GPU kernels simulate and score candidate GRN models utilizing GPU thread parallelism.
• Testing the methodology on the inference of spatiotemporal mechanistic GRNs from 2D gene expression patterns.

Main Results:

• Achieved a 700-fold speedup compared to single CPU implementations.
• Demonstrated efficient parallelization of GRN inference using GPU computing.
• Successfully inferred mechanistic GRNs capable of developing specific 2D gene expression patterns.

Conclusions:

• The proposed GPU-accelerated methodology significantly enhances the efficiency of mechanistic GRN inference.
• This approach streamlines knowledge extraction from biological data and accelerates synthetic biology applications.
• Enables the development of GRNs with specific spatial dynamics for tissue and cell culture applications.