Prediction of Eukaryotic Exons using Bidirectional LSTM-RNN based Deep Learning Model

Abstract

Exon prediction has always been a challenge for computational biologist. Although there have been many advances in identification and prediction of exons by computational methods. The efficacy and efficiency of prediction methods need to be further improved using new parameters and algorithms. Moreover, it is essential to develop new prediction methods by combining already existing approaches that can greatly improve prediction accuracy. A eukaryotic gene contains several exons and introns that are separated by splice site junction. It is important to accurately identify splice sites in a gene. Splice sites regions are known, but computational signal prediction is still challenging due to numerous false positives and other problems. In this paper, a novel combination of Support Vector Machine and bidirectional LSTM-RNN based Deep Learning approaches has been applied to improve the efficiency and accuracy of exon prediction. The proposed method takes into account the conventional machine learning as well as the deep learning approach on predictive accuracy of eukaryotic exons.