Protein-Ligand Scoring with Convolutional Neural Networks

TL;DR

This paper introduces CNN-based scoring functions for protein-ligand interactions that outperform traditional methods in ranking and screening, leveraging 3D structural data and deep learning.

Contribution

The paper presents a novel CNN scoring function trained on structural data, improving accuracy in pose prediction and virtual screening over existing scoring functions.

Findings

CNN scoring functions outperform AutoDock Vina in pose ranking

Deep learning captures key interaction features automatically

Improved virtual screening accuracy

Abstract

Computational approaches to drug discovery can reduce the time and cost associated with experimental assays and enable the screening of novel chemotypes. Structure-based drug design methods rely on scoring functions to rank and predict binding affinities and poses. The ever-expanding amount of protein-ligand binding and structural data enables the use of deep machine learning techniques for protein-ligand scoring. We describe convolutional neural network (CNN) scoring functions that take as input a comprehensive 3D representation of a protein-ligand interaction. A CNN scoring function automatically learns the key features of protein-ligand interactions that correlate with binding. We train and optimize our CNN scoring functions to discriminate between correct and incorrect binding poses and known binders and non-binders. We find that our CNN scoring function outperforms the AutoDock Vina scoring function when ranking poses both for pose prediction and virtual screening.