Encoded Deep Vectors for Eukaryotic Exon Prediction

Abstract

In bioinformatics, identifying protein-coding regions in genomic sequences is a vital problem. The majority of methods used to locate protein-coding regions(exons) in genomic sequences rely on the 3-base periodicity signal. Additionally based on encoding also, many machine learning approaches have been devised for exon prediction. They transform a sequence of DNA into numerical values and use those values to predict protein-coding regions using a machine learning model. Encoding strategies, however, have a direct impact on the classifier’s capacity to extract coding information, and it is yet unclear how to select the best encoding scheme. In this article, we proposed a hybrid encoding scheme, where the DNA nucleotide sequences are encoded into multiple vectors that were fed as multi-dimensional input channels to the Convolutional neural network. The effectiveness of the proposed hybrid encoding scheme using a Convolutional neural network is compared with the existing methods from the literature. The presented approach performed better than the existing approaches on benchmark datasets of the eukaryotic organisms, H.sapiens, D.melanogaster, C.elegans, A.thaliana, Cow, and Rat.

Supported by organization x.