DeepInterface: Protein-protein interface validation using 3D Convolutional Neural Networks

Motivation Protein–protein interactions are crucial in almost all biological processes. Proteins interact through their interfaces. It is important to determine how proteins interact through interfaces to understand protein binding mechanisms and to predict new protein-protein interactions. Results We present DeepInterface, a deep learning based method which predicts, for a given protein complex, if the interface between the proteins of a complex is a true interface or not. The model is a 3-dimensional convolutional neural networks model and the positive datasets are obtained from all complexes in the Protein Data Bank, the negative datasets are the incorrect solutions of the docking decoys. The model analyzes a given interface structure and outputs the probability of the given structure being an interface. The accuracy of the model for several interface data sets, including PIFACE, PPI4DOCK, DOCKGROUND is approximately 88% in the validation dataset and 75% in the test dataset. The method can be used to improve the accuracy of template based PPI predictions.