Accelerating Drug Discovery in AutoDock-GPU with Tensor Cores

TL;DR

A method to accelerate the sum reduction of four-element vectors using matrix operations on NVIDIA Tensor Cores using matrix operations on NVIDIA Tensor Cores is developed.

Abstract

In drug discovery, molecular docking aims at characterizing the binding of a drug-like molecule to a macromolecule. AutoDock-GPU, a state-of-the-art docking software, estimates the geometrical conformation of a docked ligand-protein complex by minimizing a scoring function. Our profiling results indicate that the current reduction operation that is heavily used in the scoring function is sub-optimal. Thus, we developed a method to accelerate the sum reduction of four-element vectors using matrix operations on NVIDIA Tensor Cores. We integrated the new reduction operation into AutoDock-GPU and evaluated it on multiple chemical complexes on three GPUs. Our results show that our method for reduction operation is 4-7 times faster than the AutoDock-GPU baseline. We also evaluated the impact of our method on the overall simulation time in the real-world docking simulation and achieved a 27% improvement on the average docking time.

Figures (8)

• Figure 1: Profiling results of a docking process of the 7cpa protein-ligand complex.
• Figure 2: The kernel launch timeline for iterations of the optimization process.
• Figure 3: Profiling results of the kernel.
• Figure 4: Distribution of average best energy values for five protein-ligand complexes using the original code, and our method.
• Figure 5: Average runtime of the two versions of the test reduction kernel on three generations of NVIDIA GPUs: T4, A100, and V100.