Improved Artificial Fish School Search Based Deep Convolutional Neural Network for Prediction of Protein Stability upon Double Mutation

Abstract

AbstractRecent advances in molecular mechanism and pharmaceutical drug designing greatly influence the process of predicting protein’s stability upon mutation by understanding the effects of amino acid substitutions. Thus, researchers started focusing on protein engineering for investigating the protein structure to discover the disease-causing variants at its initial stage and guide for designing pharmaceutical drugs. Though, there are many existing protein stability prediction models are available, still the issues related to the uncertainty and understanding in depth knowledge about the protein and molecular structure are in existence. Hence it is very challenging to construct a novel prediction model to handle the existing issues. This paper aims to develop metaheuristic based Deep Neural Network model for improved accurate prediction of protein stability. This research work proposed Fish School Search Improved Deep Convolutional Neural Network improved (CNN-FSS) for predicating protein stability upon double mutation. The parameters in CNN is potentially handled by Fish School Searching behaviour to assign best values to influence accurate prediction in presence of uncertainty and inconsistent structure of protein stability energy change in double mutation site. The complex pattern of double mutation is precisely handled by this proposed CNN-FSS, by improving the learning rate of the CNN. Instead of greedy-descent based weight assignment, the fish schooling induces optimized result. The simulation is conducted on ProTherm database and results also proved that CNN-FSS achieves highest accurate prediction of protein stability upon double mutation compared with conventional CNN and DNN.KeywordsDouble mutationProTherm databaseProtein stabilityConvolutional neural networkFish school searchEnergy changeDeep neural network