Protein Secondary Structure Prediction using Multi-input Convolutional Neural Network

Protein secondary structure prediction is a difficult problem to solve since there are no clear boundaries between protein features and coil states. Moreover, it is financially expensive and takes a long time to perform manual experiments on each protein in the lab. Predicting protein structures computationally is promising and practical. Convolutional neural network is an efficient method for learning sophisticated features since the complexity does not increase with performance. In this paper, we have constructed deep convolutional neural networks for protein inconsequential structure illusion from its dominant structure. The proposed approach attains better results with 1% higher accuracy by using primary structure as an input, and about 2% higher accuracy by adding one information of protein property, outperforming the existing solutions.