Multi-view biomedical foundation models for molecule-target and property prediction

TL;DR

This paper introduces MMELON, a multi-view foundation model integrating graph, image, and text representations for molecules, achieving robust performance across diverse biomedical tasks and aiding drug discovery.

Contribution

The paper develops a novel multi-view molecular embedding approach that combines multiple representations in a foundation model, enhancing robustness and applicability in biomedical research.

Findings

Multi-view model matches top single-view performance.

Validated on 120+ tasks including solubility and GPCR activity.

Identified potential Alzheimer's-related GPCR targets.

Abstract

Quality molecular representations are key to foundation model development in bio-medical research. Previous efforts have typically focused on a single representation or molecular view, which may have strengths or weaknesses on a given task. We develop Multi-view Molecular Embedding with Late Fusion (MMELON), an approach that integrates graph, image and text views in a foundation model setting and may be readily extended to additional representations. Single-view foundation models are each pre-trained on a dataset of up to 200M molecules. The multi-view model performs robustly, matching the performance of the highest-ranked single-view. It is validated on over 120 tasks, including molecular solubility, ADME properties, and activity against G Protein-Coupled receptors (GPCRs). We identify 33 GPCRs that are related to Alzheimer's disease and employ the multi-view model to select strong binders from a compound screen. Predictions are validated through structure-based modeling and identification of key binding motifs.