ENSDEEP-ICAM: ENSEMBLE DEEP LEARNING IDENTIFIER FOR IDENTIFICATION OF MUTATION TO DETECT COLORECTAL ADENOCARCINOMA PROGRESSION

Abstract

Colorectal cancer is the third highest widespread disease killed millions of human beings worldwide. Early detection of cancer can greatly improve survival rates. Cancer can be caused by changes in the genetic code within cells, which can be caused by mutations in the gene sequences. These mutations can be the result of errors during replication or recombination of the genome. Some mutations, referred to as cancer driver mutations, can result in cancer. Several computational methods have been devoted for identifying colorectal adenocarcinoma mutation. However, the existing approaches are inadequate to detect the mutated reigns accurately. In this study, we developed a novel ensemble deep learning-based predictor called EnsDeep-iCAM for targeting colorectal adenocarcinoma mutations. The proposed method use effective feature extraction algorithms namely position relative incidence matrix, statistical moment, reverse position relative incidence matrix, frequency vector, accumulative absolute position incidence vector, and reverse accumulative absolute position incidence vector to encode the biological mutated sequences. Then we trained the models by using different variants of deep learning algorithms such as long-short-term memory network (LSTM), bi-directional LSTM (BiLSTM), gated recurrent units(GRU) and the ensemble of all three models (Ens-Deep). BiLSTM outperformed the implemented deep learning and machine learning classifiers using ten-fold cross-validation(10-FCVT), independent test and self-consistency testing(SCT) respectively. The proposed model achieved an accuracy and MCC of 96% and 0.91on training dataset and 98% and 0.97 on testing dataset respectively. We believe that EnsDeep-iCAM protocol will provide valuable insights in characterization of large-scale unannotated mutated genes in particular and pharmaceutical drug discovery in general.