Colorectal Cancer Classification from Protein Sequences Using Several RNN Pre-Trained Models

Abstract

Bioinformatics is one field that can integrate health information with data mining applications in order to predict or analyze the patient’s information. One among the several diseases is colorectal cancer,which is horrible and rated the third leading disease in the area of cancer which leads to death in both men and women.In general, there are several intelligent methods to identify several kinds of problems in gene selection for predicting cancer,but there is no method to give a solution for patients who are diagnosed in the advanced stage.This motivated me to design theproposed approach in which we try to identify all the homo protein sequences and then train these sequences corresponding to one which causes colorectal cancer.InBioinformatics,a protein acts as one of the main agent or source to perform a biological function by interacting with molecules like Deoxyribonucleic acid (DNA), Ribonucleic acid (RNA). The function of a protein determines the healthy or diseased states of an organism. Protein interaction with other proteins can be visualized through a network called Protein-Protein Interaction Network (PPIN).In general classification of protein sequences is a very complex task. Deep learning techniques like CNN and RNN can be used to solve the problem.In computational bioinformatics,the classification of protein sequence plays an important role in determining accuracy.To improve the accuracy of our current model, the suggested method incorporates GRU, LSTM, RNN, and Customized LSTM into an RNN based architecture by optimizing the parameters in a two-way direction.Here we try to test all the models on sample protein sequences that are collected from TCGA and then determine the correctness of testing data and training data.