Comprehensive Review of Contrastive and Generative Self-Supervised Learning for Small Molecular Representation.

Area of Science:

• Computational chemistry
• Bioinformatics
• Artificial intelligence

Background:

• Drug discovery relies on analyzing complex molecular data.
• Contrastive and generative learning are key for molecular representation learning.
• These methods use diverse data, including molecular structure and biological networks.

Purpose of the Study:

• To systematically review contrastive and generative learning for molecular representations.
• To analyze their application in single-molecule and network-based learning.
• To discuss challenges and future directions in the field.

Main Methods:

• Surveying public databases and benchmarks.
• Analyzing contrastive learning methodologies.
• Analyzing generative learning methodologies.

Main Results:

• Representation learning enhances molecular property prediction, interaction analysis, and drug design.
• Methods are adapted for both single-molecule data and drug interaction networks.
• Foundation models are emerging for synergistic, multimodal learning.

Conclusions:

• Advanced representation learning is crucial for efficient drug discovery.
• The field is moving towards multimodal foundation models.
• This review guides researchers in bioinformatics, chemistry, and AI.