Transposable elements (TEs) are the most represented sequences occurring in eukaryotic genomes. Few methods provide the classification of these sequences into deeper levels, such as superfamily level, which could provide useful and detailed information about these sequences. Most methods that classify TE sequences use handcrafted features such as k-mers and homology-based search, which could be inefficient for classifying non-homologous sequences. Here we propose an approach, called transposable elements pepresentation learner (TERL), that preprocesses and transforms one-dimensional sequences into two-dimensional space data (i.e., image-like data of the sequences) and apply it to deep convolutional neural networks. This classification method tries to learn the best representation of the input data to classify it correctly. We have conducted six experiments to test the performance of TERL against other methods. Our approach obtained macro mean accuracies and F1-score of 96.4% and 85.8% for superfamilies and 95.7% and 91.5% for the order sequences from RepBase, respectively. We have also obtained macro mean accuracies and F1-score of 95.0% and 70.6% for sequences from seven databases into superfamily level and 89.3% and 73.9% for the order level, respectively. We surpassed accuracy, recall and specificity obtained by other methods on the experiment with the classification of order level sequences from seven databases and surpassed by far the time elapsed of any other method for all experiments. Therefore, TERL can learn how to predict any hierarchical level of the TEs classification system and is about 20 times and three orders of magnitude faster than TEclass and PASTEC, respectively https://github.com/muriloHoracio/TERL. Contact:murilocruz@alunos.utfpr.edu.br

Nowadays, there is an abundance of biomedical data, such as images and genetic sequences, among others. However, there is a lack of annotation to such volume of data, due to the high costs involved to perform this task. Thus, it is mandatory to develop techniques to ease the burden of human annotation. To reach such goal active learning strategies can be applied. However, the state-of-the-art active learning methods, generally, are not feasible to lead with real-world datasets. Another important issue, that is generally neglected by these methods, is related to the conception that the classifier tends to learn more and more at each iteration. Their adopted selection criteria do not properly exploit the knowledge of the classifier. Therefore, in this paper, we propose the use of an active learning approach, in order to leverage the learning process, including the proposal of a novel active learning strategy. The main difference of our proposed strategy is related to the participation of the classifier in an extremely active way in its learning process. So, we can better maximize and prioritize the knowledge that is obtained by the classifier at each iteration, making use of this knowledge in a more appropriate and useful way when selecting more informative samples. To do so, in our selection criteria, we give significant importance to the classifications suggested by the classifier. In addition, jointly with the participation and the knowledge of the classifier, we consider both uncertainty and representativeness criteria through a fine-grained analysis of the samples. Experimental results show that our novel active learning approach outperforms state-of-the-art active learning methods, considering several supervised classifiers. Hence, dealing with real dataset problems in a better way, equalizing the tradeoff between annotation task and higher accuracy rates.