The field of molecular representation learning is rapidly evolving, with a focus on developing innovative methods for aligning chemical and textual representations, predicting drug-target interactions, and improving molecular property prediction. Recent developments have highlighted the importance of incorporating multi-level protein structures, bond-centered molecular fingerprints, and flexible 2D and 3D modalities into molecular representation learning frameworks. These advancements have the potential to significantly improve the accuracy and efficiency of molecular property prediction, drug discovery, and precision medicine. Notable papers in this area include:

• Thin Bridges for Drug Text Alignment, which introduces a lightweight contrastive learning approach for aligning chemical and textual representations.
• Attending on Multilevel Structure of Proteins enables Accurate Prediction of Cold-Start Drug-Target Interactions, which proposes a framework for attending on protein multi-level structure for cold-start DTI prediction.
• Unified Molecule Pre-training with Flexible 2D and 3D Modalities, which presents a flexible molecule pre-training framework that learns unified molecular representations while supporting single-modality input.