GPU acceleration of a production molecular docking code

Abstract

Modeling the interactions of biological molecules, or docking, is critical to both understanding basic life processes and to designing new drugs. Here we describe the GPU-based acceleration of a recently developed, complex, production docking code. We show how the various functions can be mapped to the GPU and present numerous optimizations. We find which parts of the problem domain are best suited to the different correlation methods. The GPU-accelerated system achieves a speedup of at least 16x for all likely problems sizes. This makes it competitive with FPGA-based systems for small molecule docking, and superior for protein-protein docking.

A declarative concurrent system for protein structure prediction on GPU

“…Finally, we would like to mention that the use of GPU architectures has shown its advantages in several other bioinformatics applications, such as computational proteomics (Hussong, Gregorius, Tholey, & Hildebrandt, 2009), DNA sequencing (Schatz, Trapnell, Delcher, & Varshney, 2007) and molecular docking (Sukhwani & Herbordt, 2009), as well as in the context of CP and SAT solving (Campeotto, Dovier, Fioretto, & Pontelli, 2014;Dal Palù, Dovier, Formisano, & Pontelli, 2014).…”

Parameter Sweep Workflows for Modelling Carbohydrate Recognition

“…Roh et al [30] and Sukhwani and Herbordt [36], for example, both propose a molecular docking system using parallel GPUs by breaking down the computation for processing an equation as a single work unit. The performance results show that for smaller scale docking simulations the GPU based system can provide the required speed-up when compared to standalone solutions.…”

Fast binding site mapping using GPUs and CUDA

“…Though FFT reduces the computational complexity from O(N 6 ) to O(N 3 logN), our prior work on accelerating PIPER using FPGAs [15] and GPUs [16] indicates that, if the ligand grid is smaller than a certain size, direct correlation can perform better than FFT correlation, especially if multiple correlations are to be performed. This is due to many reasons: direct correlation lends itself well to parallelization, multiple correlation scores can be computed together, multiple rotations can be scored in a single pass of the protein grid and large data reuse amortizes the overhead of data fetch and kernel launch.…”

“…In our previous work, we have published acceleration of a rigid docking program using GPUs [16] and preliminary results on the acceleration of electrostatics energy computation for energy minimization [17]. Here, we extend the acceleration of energy minimization to include the van der Waals energy evaluation on GPUs.…”