Deep Learning in Drug Design: Protein-Ligand Binding Affinity Prediction.

Area of Science:

• Computational chemistry
• Structural biology
• Artificial intelligence in drug discovery

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for effective drug discovery.
• Current computational methods often require extensive feature engineering and may struggle with limited data.

Purpose of the Study:

• To introduce DeepAtom, a data-driven framework for accurate protein-ligand binding affinity prediction.
• To develop a computationally efficient model that minimizes reliance on manual feature engineering.

Main Methods:

• Utilized a 3D Convolutional Neural Network (3D-CNN) architecture within the DeepAtom framework.
• Trained and validated the model on benchmark datasets including PDBbind v.2016 and the Astex Diverse Set.
• Compiled and proposed a novel benchmark dataset to enhance model performance.

Main Results:

• DeepAtom demonstrated superior performance compared to baseline scoring methods on independent datasets.
• Achieved a Pearson's R of 0.83 and RMSE of 1.23 pK units on the PDBbind v.2016 core set using the new dataset.
• The lightweight model design showed improved representational capacity even with limited training data.

Conclusions:

• DeepAtom offers a robust and accurate solution for predicting protein-ligand binding affinity.
• The framework shows potential for integration into computational drug development workflows like molecular docking and virtual screening.
• The developed dataset and model contribute to advancing data-driven approaches in drug discovery.