LASSIE: simulating large-scale models of biochemical systems on GPUs

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Mathematical modeling and in silico analysis are crucial for understanding cellular processes.
• Simulating large-scale biological models (hundreds/thousands of reactions) is computationally intensive for Central Processing Units (CPUs).
• High-performance computing solutions like Graphics Processing Units (GPUs) can reduce computational costs.

Purpose of the Study:

• To develop a high-performance computing solution for simulating large-scale biological models.
• To reduce the computational cost and time required for in silico analysis of complex cellular processes.

Main Methods:

• LASSIE, a "black-box" GPU-accelerated deterministic simulator, was developed.
• It automatically generates Ordinary Differential Equations (ODEs) from reaction-based models using mass-action kinetics.
• Numerical solutions employ adaptive switching between Runge-Kutta-Fehlberg and Backward Differentiation Formulae based on stiffness.

Main Results:

• LASSIE achieved up to a 92x speed-up compared to a CPU-based LSODA implementation.
• Simulation time for models with thousands of reactions/species was reduced from approximately 1 month to 8 hours.
• LASSIE successfully simulated larger models than the tested CPU implementation due to a smaller memory footprint.

Conclusions:

• LASSIE utilizes a novel fine-grained parallelization strategy for efficient GPU computation.
• This GPU acceleration significantly enhances the speed and feasibility of large-scale model simulations.
• LASSIE is poised to advance Systems Biology by enabling faster, in-depth computational analyses of complex biological systems.