Scoring Functions for Protein-Ligand Binding Affinity Prediction using Structure-Based Deep Learning: A Review.

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence

Background:

• Accurate prediction of protein-ligand binding affinities is crucial for drug discovery.
• Deep learning (DL) methods are increasingly used for structure-based binding affinity prediction.
• DL-based scoring functions can outperform classical methods within their domain.

Purpose of the Study:

• To review deep learning-based structure-based scoring functions for binding affinity prediction.
• To focus on various aspects including architectures, featurization, datasets, training, and evaluation.
• To highlight the role of explainable artificial intelligence (XAI) in drug discovery.

Main Methods:

• Review of deep learning architectures for structure-based scoring functions.
• Analysis of featurization strategies for protein-ligand complexes.
• Examination of datasets, training, and evaluation methodologies.
• Discussion on the application of explainable artificial intelligence (XAI).

Main Results:

• Deep learning models demonstrate strong performance in predicting binding affinities.
• Structure-based DL methods offer advantages over traditional scoring functions.
• Explainable AI is essential for developing reliable models for drug discovery.

Conclusions:

• Deep learning significantly advances structure-based binding affinity prediction.
• The integration of XAI enhances the utility of DL models in drug discovery.
• These advancements hold transformative potential for accelerating the drug discovery pipeline.