OnionNet: a Multiple-Layer Intermolecular-Contact-Based Convolutional Neural Network for Protein-Ligand Binding...

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Accurate prediction of binding affinity is crucial for lead molecule screening in drug discovery.
• Current scoring functions often lack satisfactory accuracy for predicting protein-ligand binding.
• Machine learning and deep learning methods show promise for improving scoring function performance.

Purpose of the Study:

• To introduce OnionNet, a deep convolutional neural network for predicting binding affinity.
• To evaluate OnionNet's prediction power against existing scoring functions.
• To assess the robustness of OnionNet using docking-generated complexes.

Main Methods:

• Development of OnionNet, a deep convolutional neural network model.
• Feature engineering based on rotation-free element-pair-specific contacts between ligands and protein atoms.
• Grouping contacts into distance ranges to capture local and nonlocal interactions.
• Evaluation using the CASF-2013 benchmark and PDBbind v2016 core set.
• Testing robustness on docking-generated protein-ligand complexes.

Main Results:

• OnionNet demonstrates competitive or superior prediction power compared to other scoring functions.
• The model effectively captures both local and nonlocal interaction information.
• OnionNet shows robustness when applied to complexes from docking simulations.

Conclusions:

• OnionNet represents an advancement in computational drug discovery for binding affinity prediction.
• The model's feature representation and deep learning architecture enhance accuracy.
• OnionNet offers a promising tool for large-scale lead molecule screening.