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

Shudong Wang,Mao Ding,Jinfu Zhu,Renteng Zhao,Zhenzhen Du,Tao Song,Dayan Liu,Yue Zhong

doi:10.3389/fgene.2020.607824

Abstract

Deep learning methods, which can predict the binding affinity of a drug–target protein interaction, reduce the time and cost of drug discovery. In this study, we propose a novel deep convolutional neural network called SE-OnionNet, with two squeeze-and-excitation (SE) modules, to computationally predict the binding affinity of a protein–ligand complex. The OnionNet is used to extract a feature map from the three-dimensional structure of a protein–drug molecular complex. The SE module is added to the second and third convolutional layers to improve the non-linear expression of the network to improve model performance. Three different optimizers, stochastic gradient descent (SGD), Adam, and Adagrad, were also used to improve the performance of the model. A majority of protein–molecule complexes were used for training, and the comparative assessment of scoring functions (CASF-2016) was used as the benchmark. Experimental results show that our model performs better than OnionNet, Pafnucy, and AutoDock Vina. Finally, we chose the macrophage migration inhibitor factor (PDB ID: 6cbg) to test the stability and robustness of the model. We found that the prediction results were not affected by the docking position, and thus, our model is of acceptable robustness.

Highlights

The binding affinity of small molecules to receptor proteins is the key to drug discovery and drug repositioning (David Hecht, 2009; Ru et al, 2020; Zeng et al, 2020a)
We found that Adagrad was the fastest optimizing algorithm with an accuracy higher than that of stochastic gradient descent (SGD) and Adam
A modified deep learning model SE-OnionNet, with an attention mechanism to improve the performance of the model, is constructed

Summary

Introduction

The binding affinity of small molecules to receptor proteins is the key to drug discovery and drug repositioning (David Hecht, 2009; Ru et al, 2020; Zeng et al, 2020a). The development of accurate prediction models for calculating binding affinity is imperative. The OnionNet model (Zheng et al, 2019) was proposed for predicting binding affinity using the three-dimensional structure of complexes. In the search for a favorable docking pose, a specific scoring function is used to estimate the binding affinity often with a low accuracy and a high false-positive rate. The molecular mechanics Poisson–Boltzmann surface area method (Rd et al, 2012) was developed to calculate the binding free energy. This method is computationally intensive and is generally superior to the docking scoring function (Shoichet, 2004)

Methods

Results

Discussion

Conclusion