Advances in Protein-Ligand Binding Affinity Prediction via Deep Learning: A Comprehensive Study of Datasets, Data...

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery

Background:

• Drug discovery is complex and costly, with lead compound identification being a critical phase.
• Computational methods, particularly deep learning (DL), are increasingly used to predict protein-ligand interactions and binding affinity.
• Existing research lacks a comprehensive analysis of datasets and recent DL methods for binding affinity prediction (BAP).

Purpose of the Study:

• To provide a comprehensive survey of commonly used datasets for BAP, discussing their quality and limitations.
• To classify and analyze recent DL methods applied to BAP.
• To offer a fresh perspective on the evolving field of DL in BAP.

Main Methods:

• Systematic examination of datasets commonly used for BAP, including their characteristics and preprocessing steps.
• Review of various DL techniques such as graph neural networks, convolutional neural networks, and transformers for BAP.
• Extensive literature research to include the most recent DL approaches for BAP.

Main Results:

• Identified inherent challenges in DL-based BAP, including data quality, model interpretability, and explainability.
• Highlighted key considerations for future research directions in DL for BAP.
• Provided valuable insights to accelerate the development of effective and reliable DL models for BAP.

Conclusions:

• The study offers a comprehensive overview that can significantly enhance future research in predicting protein-ligand binding affinity.
• Improved BAP models can accelerate the identification of lead compounds, thereby streamlining the overall drug development process.