An Improved Promoter Recognition Model Using Convolutional Neural Network

Abstract

Gene expression is regulated by transcription and translation, and the promoter controls the start of transcription. Finding the exact location of a promoter is of great importance to life science. With the development of the Next-Generation Sequencing (NGS), more and more eukaryotic gene sequence data are available. Computational prediction of eukaryotic promoters has become one of the most challenging problems in sequence analysis. Many methods have been proposed, but the accuracy of prediction still needs to be improved. In this paper we use support vector machine (SVM) to verify that promoter elements are more important than non-elements for predicting promoters. With this factor in mind, we utilize convolution filters to compress non-elements information, and encode elements to emphasize their importance. A new prediction model is constructed based on neural networks. We applied a 10-fold cross validation test to validate the proposed model. We achieved 89.86% accuracy, 86.51% specificity and 89.64% sensitivity, which are better than the other three prediction methods (SVM, NNPP2.2 and CNN).