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

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Reverse engineering gene regulatory networks (GRNs) with spatial dynamics is challenging due to computational intensity.
• Existing heuristic machine learning algorithms require extensive simulations, limiting their practical application.
• There is a need for efficient methods to infer GRNs from spatial gene expression patterns, especially utilizing parallel computing.

Purpose of the Study:

• To develop an efficient methodology for reverse engineering mechanistic GRNs using GPU computing.
• To accelerate the inference of GRNs capable of generating specific spatial gene expression patterns.
• To enable faster extraction of biological knowledge and aid synthetic biology design.

Main Methods:

• Parallelization of evolutionary algorithms using Graphics Processing Unit (GPU) computing.
• Asynchronous launching of GPU kernels for parallel simulation and scoring of candidate GRN models.
• Multi-CPU threads managing evolutionary operators (crossover, mutation, selection) while GPUs handle model simulations.

Main Results:

• Achieved a 700-fold speedup compared to single CPU implementations for GRN inference.
• Successfully inferred spatiotemporal mechanistic GRNs, including topology and parameters, from 2D gene expression patterns.
• Demonstrated the efficiency and scalability of the GPU-accelerated approach.

Conclusions:

• The proposed GPU-based methodology significantly accelerates the reverse engineering of mechanistic GRNs from spatial data.
• This approach overcomes computational limitations, making complex GRN inference more accessible.
• Facilitates advancements in understanding biological systems and designing synthetic gene networks.