SE-OnionNet: A Convolution Neural Network for Protein-Ligand Binding Affinity Prediction.

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Drug discovery is a time-consuming and expensive process.
• Predicting drug-target binding affinity is crucial for efficient drug development.
• Deep learning offers a promising approach to accelerate binding affinity prediction.

Purpose of the Study:

• To propose SE-OnionNet, a novel deep convolutional neural network for predicting protein-ligand binding affinity.
• To enhance the performance of existing deep learning models for drug discovery.
• To improve the accuracy and robustness of computational binding affinity predictions.

Main Methods:

• Utilized a deep convolutional neural network (SE-OnionNet) incorporating two squeeze-and-excitation (SE) modules.
• Employed the OnionNet architecture for feature map extraction from 3D protein-drug complexes.
• Integrated SE modules into convolutional layers to boost non-linear expression and model performance.
• Experimented with SGD, Adam, and Adagrad optimizers.
• Trained the model on a large dataset of protein-molecule complexes and benchmarked against CASF-2016.

Main Results:

• SE-OnionNet demonstrated superior performance compared to OnionNet, Pafnucy, and AutoDock Vina.
• The addition of SE modules improved the network's non-linear expression and overall performance.
• Model robustness was confirmed through testing on the macrophage migration inhibitor factor (6cbg), showing independence from docking position.

Conclusions:

• SE-OnionNet is an effective deep learning model for predicting protein-ligand binding affinity.
• The model significantly enhances the accuracy and robustness of computational drug discovery.
• SE-OnionNet represents a valuable tool for accelerating the identification of potential drug candidates.