Efficient molecular encoders for virtual screening.

Area of Science:

• Computational chemistry
• Cheminformatics
• Artificial intelligence in drug discovery

Background:

• Molecular representations are crucial for effective molecular virtual screening (VS).
• Numerous molecular encoders have been developed and evaluated in VS challenges.
• Deep learning (DL) based encoders offer automatic feature extraction capabilities.

Purpose of the Study:

• To provide an overview of 2D, 3D, and DL-based molecular encoders.
• To summarize recent advancements in VS utilizing DL technologies.
• To propose a general framework for DL molecular encoder-based VS.

Main Methods:

• Review of existing literature on molecular encoders (2D, 3D, DL).
• Analysis of recent progress in DL-driven virtual screening.
• Development of a conceptual framework for DL-based VS.

Main Results:

• DL-based molecular encoders demonstrate significant potential for automatic information extraction.
• Combinational strategies can enhance VS performance.
• The review synthesizes current knowledge and proposes a unified framework.

Conclusions:

• DL molecular encoders represent a promising direction for improving VS.
• Future research should focus on advancing molecular representations and their applications in predicting active compounds.
• The proposed framework can guide future development in DL-based VS.