Recent progress on the prospective application of machine learning to structure-based virtual screening.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Machine learning (ML) is increasingly used to develop accurate scoring functions (SFs) by leveraging vast bioactivity and protein structure data.
• Advances in training and evaluation methodologies are enhancing the performance and applicability of ML-based SFs.

Purpose of the Study:

• To review recent advances in ML-based SFs, focusing on decoy selection strategies.
• To examine the application of ML-based SFs in prospective structure-based virtual screening (SBVS).
• To provide recommendations for future SBVS studies.

Main Methods:

• Review of recent literature on ML-based SFs, decoy selection, and SBVS.
• Analysis of improvements in ML-based SFs compared to classical methods in SBVS.

Main Results:

• ML-based SFs show improved accuracy and broader applicability due to better training and evaluation.
• Optimal decoy selection is crucial for training and testing ML-based SFs.
• ML-based SFs demonstrate superior performance in prospective SBVS compared to classical SFs.

Conclusions:

• Continued development in ML-based SFs, particularly in decoy selection, enhances their utility.
• ML-based SFs represent a significant advancement for structure-based virtual screening.
• Methodological insights guide future prospective SBVS studies.