Data-parallel algorithms for agent-based model simulation of tuberculosis on graphics processing units

Abstract

Agent-based modeling has been recognized as a method to bridge the translational gap in integrative systems biology. However, the computational complexity of agent-based models at biologically relevant scales makes simulation impractical on traditional CPU-based serial computing. In this paper we present a series of algorithms for simulating large scale agent-based models on graphics processing units (GPUs). GPUs have recently emerged as a powerful and economical computing platform for certain applications in scientific computing. As a test case, we have implemented an agent-based model of tuberculosis. This model simulates the interaction of the human immune system in the lung with Mycobacterium tuberculosis and tracks the formation of characteristic structures called granulomas. The model uses mobile agents to represent immune cells such as T cells and macrophages, field equations representing effector chemokines, and bacteria. Algorithms were implemented and benchmarked against a CPU implementation. Our benchmarks show performance gains of over 100 for moderately sized models. This opens the possibility of efficiently simulating realistically sized models on desktop computers.