OnionNet-2: A Convolutional Neural Network Model for Predicting Protein-Ligand Binding Affinity Based on Residue-Atom...

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate prediction of protein-ligand binding affinity (△G) is crucial for virtual screening.
• Deep learning (DL) models show promise in enhancing scoring function accuracy by extracting complex features.
• Further improvements in prediction accuracy and computational efficiency are needed.

Purpose of the Study:

• To develop a simple and efficient scoring function for predicting protein-ligand binding free energy.
• To introduce OnionNet-2, a convolutional neural network-based model for △G prediction.
• To validate the model's performance against established benchmarks and diverse datasets.

Main Methods:

• Utilized a convolutional neural network architecture (OnionNet-2).
• Characterized protein-ligand interactions by counting atom-residue contacts in multiple distance shells.
• Evaluated the model on CASF-2016, CASF-2013, decoy structures, and the CSAR NRC-HiQ dataset.

Main Results:

• OnionNet-2 demonstrated superior performance compared to existing models on CASF-2016 and CASF-2013 benchmarks.
• The model achieved significant success in predicting binding affinity for decoy structures and the high-quality CSAR NRC-HiQ dataset.
• The proposed method provides a simple yet effective approach for △G prediction.

Conclusions:

• OnionNet-2 is a highly effective scoring function for predicting protein-ligand binding free energy.
• The model's simplicity and efficiency make it a valuable tool for virtual screening and drug discovery.
• This work contributes a robust and computationally feasible solution to a key challenge in computational chemistry.