Boosting predictions of Host-Pathogen protein interactions using Deep neural networks

Abstract

The initiation of the infection process in a living organism starts with the interaction of host protein with the pathogen protein. So, the prediction of this host-pathogen protein interaction (HPI) can help in drug design and disease management strategy. Investigation of HPI by high-throughput experimental techniques is expensive and time-consuming. Therefore computational techniques have come up as an effective alternative for the prediction of these interactions. In this paper, a Deep neural network-based HPI prediction model is proposed. In the proposed technique first, the variable-length protein sequences are encoded into fixed-length input by using a Local descriptor based feature extraction method. These features are used as input to DNN based predictor. An exhaustive simulation study shows 91.70% and 87.30% accuracy on Human- Bacillus Anthracis and Human- Yersinia pestis datasets.