Accelerating a Geometric Approach to Molecular Docking with OpenACC

Abstract

In a drug discovery process, the Molecular Docking task aims at estimating the three-dimensional pose of a molecule when it interacts with the target protein. This task is usually used to perform a screening on a large library of molecules to find the most promising candidates. The output of this task is used to estimate the actual strength of atomic interactions. In this document we focus on an application that performs molecular docking using geometrical features of the molecule and of the protein, to quickly screen the target chemical library.Due to the size of the chemical library and to the complexity of the task, the application is a typical batch job that runs in an HPC platform, optimized for CPU processing. Given the amount of parallelism of this application, we evaluate the possibility to run such application on a GPU node, leveraging the OpenACC directive language.Preliminary results show that we are able to achieve a significant speedup on the kernel that was the bottleneck on the CPU (up to 16x), while we achieve a modest speedup on the overall execution (5x).

Exploiting OpenMP and OpenACC to accelerate a geometric approach to molecular docking in heterogeneous HPC nodes

“…This paper extends our previous work [17], where we developed a pure GPU version of the algorithm, where both the main kernels (alignment and pose optimization) are offloaded to the GPU. In particular, starting from the profiling analysis, we implemented a first version of the algorithm that aims at minimizing the data transfer while maintaining the application structure.…”

“…Despite the multiplatform approach of those languages, the kernels must be tuned according to the target platform, since performance portability is still an open problem [12]. Indeed, one of the conclusions of our previous work [17] was that the code modified and optimized for the GPU was not efficient when compiled and run on the CPU as the original code.…”

“…From experimental results, the best size configuration for each function is: This information confirmed us the hypothesis we made beforehand: the second kernel is less GPU-friendly than the first one. More in-depth analyses of this version are described in previous work [17].…”

“…In our previous work [17], we investigated the benefits and limitations of using the OpenACC [4] language extension in a molecular docking application, to mitigate the second problem. In this work, we implemented a hybrid version by using OpenMP [2] and OpenACC to leverage the processing elements on a heterogeneous node.…”

A Review on Parallel Virtual Screening Softwares for High-Performance Computers

“…LiGen [60] is a VS software that leverages CPU and GPU to perform the required computation. Several versions of the tool have been developed, starting from a CPU-only application [20,60,61], then the main kernels have been ported to the GPU by [62] using OpenACC. Finally, it has been optimized using CUDA for the GPU kernels and this last version has been used to perform a large VS experiment in the search of a therapeutic cure for COVID-19 screening 71.6 B compounds against 15 binding sites from 12 Sars-Cov2 viral proteins on two supercomputers, accounting for 81PFlops [41].…”