Structure-based, deep-learning models for protein-ligand binding affinity prediction.

Area of Science:

• Molecular structural science
• Computational chemistry
• Biophysics

Background:

• Deep learning has advanced molecular structural science, exemplified by the AlphaFold series.
• Predicting protein-ligand binding affinity is a critical challenge requiring sophisticated computational methods.

Purpose of the Study:

• To review mainstream structure-based deep learning approaches for protein-ligand binding affinity prediction.
• To assess the readiness of deep learning for this complex problem.
• To provide a uniform basis for comparing different deep learning models.

Main Methods:

• Focus on molecular representations, learning architectures, and model interpretability in deep learning models.
• Generated a taxonomy of existing deep learning models.
• Evaluated representative models on a uniform basis to enable valid comparisons.

Main Results:

• Identified and categorized mainstream structure-based deep learning approaches.
• Provided a comparative analysis of model advantages and shortcomings.
• Highlighted the need for standardized evaluation protocols.

Conclusions:

• Deep learning holds significant potential for advancing protein-ligand binding affinity prediction.
• The review offers a framework for understanding and comparing deep learning models in this field.
• Findings can benefit structure-based drug discovery and related research areas.