Machine learning and ligand binding predictions: A review of data, methods, and obstacles

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Accurate ligand binding prediction is crucial but computationally expensive.
• Machine learning (ML) offers a promising alternative by leveraging big biomedical data.
• Current ML models for drug binding prediction face challenges with generalizability.

Purpose of the Study:

• To review current trends in ML for drug binding prediction.
• To identify data sources for developing ML algorithms.
• To address potential issues like overfitting and ungeneralizable models.

Main Methods:

• Review of current literature on ML in drug binding prediction.
• Characterization of popular datasets using spatial statistics.
• Evaluation of benchmark datasets to identify biases.

Main Results:

• Good ML model performance correlates with high predicted bias scores.
• Models with low bias scores exhibit limited predictive power.
• Spatial statistics can quantify biases in benchmark datasets.

Conclusions:

• Understanding and mitigating data biases are critical for developing generalizable ML models.
• Improved data quality and bias assessment will enhance virtual high-throughput screening.
• This research paves the way for more effective ML-driven novel drug discovery.