A Novel Framework for Protein Sequence Classification using LSTM and CNN

Abstract

Protein sequence classification has retained its status as one of the biggest challenges in the pharmaceutical industry due to the involved complexities of the subject matter. Technological developments have justified that there needs to be (some) knowledge of identifying and classifying the protein families. On the other hand, and rather cursorily, current methodologies consider a meagre set of protein sequence descriptors. The authors set forth here the amalgamation of BLSTM and BTCN wherein the deeper features within the protein sequence shall be explored. The thesis seeks to divide protein sequences by the sliding duration of the sliding window applied. Local dependencies within these segments may be explored with the proposed convolutional network, capturing interactions between global residues involving BLSTM network. Hence, a BLSTM has been taken as needed in the whole creation, because there is a dependency between amino acid classification and past and future secondary features, and method is that it achieves bidirectional properties avoiding any knowledge of the past and future information for gaining a necessary amount of additional insight. The proposed ensemble model has been concluded as more suitable for protein structure prediction research