Advancing Drug Discovery with Enhanced Chemical Understanding via Asymmetric Contrastive Multimodal Learning

2023-11-11

AI Summary

Molecular representation learning in drug discovery suffers from insufficient semantic understanding and inefficient multimodal fusion. Method: We propose Asymmetric Contrastive Multimodal Learning (ACML), a framework that jointly models SMILES, InChI, molecular images, and graph structures. It employs pretrained unimodal encoders alongside a lightweight 5-layer GNN, leveraging an asymmetric contrastive loss to efficiently transfer cross-modal semantics into graph representations. The design balances expressiveness, interpretability, and training efficiency. Contribution/Results: ACML achieves state-of-the-art performance across cross-modal retrieval, stereoisomer discrimination, and molecular property prediction on MoleculeNet and Therapeutics Data Commons (TDC) benchmarks. Moreover, it enhances the chemical interpretability of GNNs and improves cross-dataset generalization. By enabling end-to-end, multimodally grounded chemical understanding, ACML establishes a novel paradigm for data-driven drug discovery.

Abstract

The versatility of multimodal deep learning holds tremendous promise for advancing scientific research and practical applications. As this field continues to evolve, the collective power of cross-modal analysis promises to drive transformative innovations, opening new frontiers in chemical understanding and drug discovery. Hence, we introduce Asymmetric Contrastive Multimodal Learning (ACML), a specifically designed approach to enhance molecular understanding and accelerate advancements in drug discovery. ACML harnesses the power of effective asymmetric contrastive learning to seamlessly transfer information from various chemical modalities to molecular graph representations. By combining pre-trained chemical unimodal encoders and a shallow-designed graph encoder with 5 layers, ACML facilitates the assimilation of coordinated chemical semantics from different modalities, leading to comprehensive representation learning with efficient training. We demonstrate the effectiveness of this framework through large-scale cross-modality retrieval and isomer discrimination tasks. Additionally, ACML enhances interpretability by revealing chemical semantics in graph presentations and bolsters the expressive power of graph neural networks, as evidenced by improved performance in molecular property prediction tasks from MoleculeNet and Therapeutics Data Commons (TDC). Ultimately, ACML exemplifies its potential to revolutionize molecular representational learning, offering deeper insights into the chemical semantics of diverse modalities and paving the way for groundbreaking advancements in chemical research and drug discovery.

Problem

• Enhancing molecular understanding
• Accelerating drug discovery
• Improving chemical representation learning

Innovation

• Asymmetric Contrastive Multimodal Learning
• Cross-modal chemical semantics assimilation
• Enhanced graph neural network interpretability