Abstract:This study introduces a novel approach to enhance molecular representation learning by integrating high‐content cell microscopy images at the phenotypic level. The proposed unified framework employs contrastive loss functions for cross‐modal pre‐training, enabling mutual retrieval of molecules and images. The model improves not only molecular properties predictions but also clinical outcome predictions, highlighting the potential of cross‐modality learning in bridging molecular structures to phenotypes for drug discovery. Constructing discriminative representations of molecules lies at the core of a number of domains such as drug discovery, chemistry, and medicine. State‐of‐the‐art methods employ graph neural networks and self‐supervised learning (SSL) to learn unlabeled data for structural representations, which can then be fine‐tuned for downstream tasks. Albeit powerful, these methods are pre‐trained solely on molecular structures and thus often struggle with tasks involved in intricate biological processes. Here, it is proposed to assist the learning of molecular representation by using the perturbed high‐content cell microscopy images at the phenotypic level. To incorporate the cross‐modal pre‐training, a unified framework is constructed to align them through multiple types of contrastive loss functions, which is proven effective in the formulated novel tasks to retrieve the molecules and corresponding images mutually. More importantly, the model can infer functional molecules according to cellular images generated by genetic perturbations. In parallel, the proposed model can transfer non‐trivially to molecular property predictions, and has shown great improvement over clinical outcome predictions. These results suggest that such cross‐modality learning can bridge molecules and phenotype to play important roles in drug discovery.

Molecular contrastive learning of representations via graph neural networks

Improving Molecular Contrastive Learning via Faulty Negative Mitigation and Decomposed Fragment Contrast

MoCL: Data-driven Molecular Fingerprint via Knowledge-aware Contrastive Learning from Molecular Graph

Molecular Contrastive Learning with Chemical Element Knowledge Graph

Molecular Graph Contrastive Learning with Parameterized Explainable Augmentations

Attention-wise masked graph contrastive learning for predicting molecular property

3D graph contrastive learning for molecular property prediction

GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction

CasANGCL: pre-training and fine-tuning model based on cascaded attention network and graph contrastive learning for molecular property prediction

Contrastive Dual-Interaction Graph Neural Network for Molecular Property Prediction

3D-Mol: A Novel Contrastive Learning Framework for Molecular Property Prediction with 3D Information

DIG-Mol: A Contrastive Dual-Interaction Graph Neural Network for Molecular Property Prediction

Knowledge-aware Contrastive Molecular Graph Learning

MolCloze - A Unified Cloze-style Self-supervised Molecular Structure Learning Model for Chemical Property Prediction.

MoleMCL: a multi-level contrastive learning framework for molecular pre-training

Chemical-Reaction-Aware Molecule Representation Learning

Cross‐Modal Graph Contrastive Learning with Cellular Images

Asymmetric Contrastive Multimodal Learning for Advancing Chemical Understanding

SMICLR: Contrastive Learning on Multiple Molecular Representations for Semisupervised and Unsupervised Representation Learning

A 3D-Shape Similarity-based Contrastive Approach to Molecular Representation Learning