Classification of Transposable Elements by Convolutional Neural Networks

TL;DR: This work presents a method that classifies TEs by training a CNN to label them in classes, orders and superfamilies, and presents very promising results, outperforming PASTEC and REPCLASS.

Abstract: The correct classification of transposable elements (TEs) present in the genomes is crucial to understand the real role and the consequences of these elements on the organisms. Here we present a method that classifies TEs by training a CNN to label them in classes, orders and superfamilies. Unlike previous works in the literature, the proposed method does not search for similarities to classify the sequences or use traditional machine learning classifiers. Instead of that, it automatically extracts features and classify the sequences by the CNN itself. We performed an extensive experimental evaluation, analyzing our proposed method under different scenarios. It was capable to classify TEs’ sequences from various datasets in 9 different superfamilies and obtained an accuracy of \(94\%\). We also present comparisons between the proposed method and other state-of-the-art classification tools (PASTEC, REPCLASS and TECLASS), our method presents very promising results, outperforming PASTEC and REPCLASS.