Ligand Pose Ensembles Improve Affinity Prediction in Structure-Based Virtual Screening

At the American Chemical Society Fall 2020 Virtual Meeting & Expo, several Atomwise members and partners were selected to present their research and work. Learn what our Atoms have been working on below and visit Atomwise at ACS Fall 2020 Virtual Meeting & Expo for other presentation sessions. 

Kate Stafford, PhD 
Atomwise

Atomwise Collaborators: Brandon Anderson, PhD

Title: Ligand Pose Ensembles Improve Affinity Prediction in Structure-Based Virtual Screening

Division: COMP

Abstract

Accurate and efficient prediction of binding affinity between small molecules and proteins is an important component of drug discovery and lead optimization. Deep learning has emerged as a powerful tool for structure-based affinity predictions. However, many factors limit the quality of these predictions, including the challenge of making structure-based predictions from static snapshots. Here we explore the effect of incorporating ligand conformational ensembles on deep learning based affinity prediction. Using the publicly available PDBBind-2019 dataset, we show that a small ensemble of docked poses outperforms single poses in affinity prediction tasks even when the protein is treated as rigid. We explore a variety of pose pooling strategies, including mean-pooling, max-pooling, and attention layers applied to both hidden states and predictions. We conclude with a discussion of the strengths and weaknesses of each pooling technique.

Poster Presentation

On-demand Poster (click for hi-resolution)

On-demand audio recording

Join our team