Prediction of Intrinsically Disordered Proteins with Convolutional Neural Networks based on Feature Selection

Abstract

Intrinsically disordered proteins (IDPs) possess flexible 3-D structures, which make them play an important role in a variety of biological functions. We develop a method to predict intrinsically disordered proteins based on feature selection and convolutional neural networks (CNN). The combination of structural, physicochemical and evolutionary properties is used to describe the differences between disordered and ordered regions. Especially, to highlight the correlation between the target residue and adjacent residues, multiple windows are selected to preprocess the selected properties. After that, these calculated properties are combined into the feature matrix to predict IDPs through the constructed CNN. Our method is training as well as testing based on the DisProt database. The simulation results show that the proposed method can predict intrinsically disordered proteins effectively, and the performance is competitive in comparison with IsUnstruct and ESpritz.