DSDP: A Blind Docking Strategy Accelerated by GPUs

YuPeng Huang,Hong Zhang,Siyuan Jiang,Dajiong Yue,Xiaohan Lin,Jun Zhang,Yi Qin Gao

2023-03-16

Abstract:Virtual screening, including molecular docking, plays an essential role in drug discovery. Many traditional and machine-learning based methods are available to fulfil the docking task. The traditional docking methods are normally extensively time-consuming, and their performance in blind docking remains to be improved. Although the runtime of docking based on machine learning is significantly decreased, their accuracy is still limited. In this study, we take the advantage of both traditional and machine-learning based methods, and present a method Deep Site and Docking Pose (DSDP) to improve the performance of blind docking. For the traditional blind docking, the entire protein is covered by a cube, and the initial positions of ligands are randomly generated in the cube. In contract, DSDP can predict the binding site of proteins and provide an accurate searching space and initial positions for the further conformational sampling. The docking task of DSDP makes use of the score function and a similar but modified searching strategy of AutoDock Vina, accelerated by implementation in GPUs. We systematically compare its performance with the state-of-the-art methods, including Autodock Vina, GNINA, QuickVina, SMINA, and DiffDock. DSDP reaches a 29.8% top-1 success rate (RMSD < 2 Å) on an unbiased and challenging test dataset with 1.2 s wall-clock computational time per system. Its performances on DUD-E dataset and the time-split PDBBind dataset used in EquiBind, TankBind, and DiffDock are also effective, presenting a 57.2% and 41.8% top-1 success rate with 0.8 s and 1.0 s per system, respectively.

Chemical Physics,Machine Learning

What problem does this paper attempt to address?

The paper aims to address the speed and accuracy issues in molecular docking for drug discovery, especially in the context of blind docking. Traditional docking methods, while widely used, have room for improvement in efficiency and performance in blind docking tasks; machine learning-based methods, although they can significantly reduce runtime, still have limited accuracy. To tackle this issue, the paper introduces a new approach that combines traditional and machine learning strategies—Deep Site and Docking Pose (DSDP), aiming to enhance the performance of blind docking.

The main contributions of the DSDP method are:
1. Utilizing machine learning to predict protein binding sites, providing precise search space and initial positions for subsequent conformational sampling, thereby reducing the space required for blind searching across the entire protein.
2. Using a search strategy similar to but modified from AutoDock Vina, and implementing acceleration on GPUs, the docking speed is improved.
3. On a series of benchmark datasets, such as DUD-E and time-split PDBBind datasets, DSDP has shown a high success rate (top-1 success rate reaching 29.8%-41.8%), while maintaining a fast computation speed (0.8 to 1.2 seconds per system).

The paper also provides a detailed introduction to the workflow of DSDP, methodological details, and comparisons with existing technologies, demonstrating its advantages in improving the speed and accuracy of blind docking.