An explainable one-dimensional convolutional neural network with modified Gabor wavelet transform for the identification of exons.

Abstract

In this paper, we propose an effective one-dimensional CNN (1D-CNN) model for the identification of exons by considering the features extracted from the DNA sequences using DSP approaches (short-time discrete Fourier transform and the modified Gabor wavelet transform), along with various numerical mapping methods. To preserve the feature information without any information loss, a novel CNN model is proposed by excluding the pooling layer. The experimental outcomes reveal that the 1D-CNN model with the Voss-MGWT feature extraction method outperforms other discussed methods in improving the identification accuracy by using the HMR195 dataset.