A machine learning based framework to identify and classify long terminal repeat retrotransposons

Abstract

Transposable elements (TEs) are repetitive nucleotide sequences that make up a large portion of eukaryotic genomes. They can move and duplicate within a genome, increasing genome size and contributing to genetic diversity within and across species. Accurate identification and classification of TEs present in a genome is an important step towards understanding their effects on genes and their role in genome evolution. We introduce TE-Learner, a framework based on machine learning that automatically identifies TEs in a given genome and assigns a classification to them. We present an implementation of our framework towards LTR retrotransposons, a particular type of TEs characterized by having long terminal repeats (LTRs) at their boundaries. We evaluate the predictive performance of our framework on the well-annotated genomes of Drosophila melanogaster and Arabidopsis thaliana and we compare our results for three LTR retrotransposon superfamilies with the results of three widely used methods for TE identification or classification: RepeatMasker, Censor and LtrDigest. In contrast to these methods, TE-Learner is the first to incorporate machine learning techniques, outperforming these methods in terms of predictive performance, while able to learn models and make predictions efficiently. Moreover, we show that our method was able to identify TEs that none of the above method could find, and we investigated TE-Learner’s predictions which did not correspond to an official annotation. It turns out that many of these predictions are in fact strongly homologous to a known TE.

Peer Review #1 of "A systematic review of the application of machine learning in the detection and classification of transposable elements (v0.1)"

“…In recent years, much bioinformatics software has been developed to detect TEs (Girgis, 2015) and, although they follow different strategies (such as homology-based, structure-based, de novo, and using comparative genomics), these lack sensitivity and specificity due to the polymorphic structures of TEs (Su, Gu & Peterson, 2019). Loureiro et al (Loureiro et al, 2013a) proved that ML could be used to improve the accuracy of TEs detection by combining results obtained by several conventional software and training a classifier using these results (Schietgat et al, 2018), (Loureiro et al, 2013b). Loureiro's work provided novel evidence for the use of ML in TEs, yet it did not use ML to obtain the predictions, making the results too dependent on traditional algorithms.…”

“…Loureiro's work provided novel evidence for the use of ML in TEs, yet it did not use ML to obtain the predictions, making the results too dependent on traditional algorithms. Using the Random Forest algorithm, Schietgat et al were able to improve results obtained by popular bioinformatics software (which followed a homology-based strategy) such as Censor, RepeatMasker, and LTRDigest (Schietgat et al, 2018) in the detection of LTR retrotransposons. The authors proposed a framework called TE-Learner LTR , which outperformed LTRDigest in recall and RepeatMasker and Censor in terms of precision.…”

“…For example, (Loureiro et al, 2013b), (Zamith Santos et al, 2018), and (Abrusan et al, 2009) used k-mer frequencies as features. On the contrary, (Schietgat et al, 2018), (Arango-López et al, 2017), (Loureiro et al, 2013a) used numerical and categorical features mainly based on structures. Other purposes of ML in TEs included detecting the boundaries of mobile elements (Tsafnat et al, 2011), identifying insertion sites of somatic LINEs insertions (Tang et al, 2017), and other mobile elements at the genome level (Rawal & Ramaswamy, 2011), detecting aneuploidy in patients with cancer through LINEs (Douville et al, 2018), detecting conserved motifs in TIR elements (Su, Gu & Peterson, 2019), distinguishing endogenous retroviral LTRs from SINEs (Ashlock & Datta, 2012), and comparing multiple studies on transposon insertion sequencing (Hubbard et al, 2019).…”

“…PRISMA flow diagram Publication identifier Year Q1 Q2 Q3 Q4 Reference Publication identifier Year Q1 Q2 Q3 Q4 Reference P1 X X X X (Yu, Yu & Pan, 2017) P19 2013 X X X (Loureiro et al, 2013b) P2 X X X (Schietgat et al, 2018) P20 2014 X X (Ma, Zhang & Wang, 2014) P3 X X X (Arango-López et al, 2017) P21 2010 X X (Dashti & Masoudi-Nejad, 2010) P4 X X X (Loureiro et al, 2013a) P22 2010 X X (Ding, Zhou & Guan, 2010) P5 X X X (Tsafnat et al, 2011) P23 2019 X (Jaiswal & Krishnamachari, 2019) P6 X X X (Zhang et al, 2018) P24 2015 X X X X (Girgis, 2015) P7 X X X (Eraslan et al, 2019) P25 2018 X X X (Nakano et al, 2018a) P8 X X X (Douville et al, 2018) P26 2018 X X X (Zamith Santos et al, 2018) P9 X X (Chen et al, 2018) P27 2009 X (Abrusan et al, 2009) P10 X X X X (Ashlock & Datta, 2012) P28 2019 X X (Su, Gu & Peterson, 2019) P11 X X X (Smith et al, 2017) P29 2017 X X X X (Nakano et al, 2017) P12 X X X X (Kamath, De Jong & Shehu, 2014) P30 2014 X X X (Brayet et al, 2014) P13 X X X (Kim et al, 2016) P31 2013 X (Zamani et al, 2013) P14 X X X (Segal et al, 2018) P32 2019 X (Hubbard et al, 2019) P15 X X X (Rawal & Ramaswamy, 2011) P33 2014 X X (Ryvkin et al, 2014) P16 X X X (Tang et al, 2017) P34 2013 X X X X (Zhang et al, 2013) P17 X X X (Ventola et al, 2017) P35 2019 X X…”

A systematic review of the application of machine learning in the detection and classification of transposable elements

“…Due to the high diversity of TE structures and transposition mechanisms, there are still numerous classification problems and debates on the classification systems (Piégu et al, 2015). TEs in eukaryotes are traditionally classified based on if the reverse transcription is needed for transposition (Class I or retrotransposons) or not (Class II or DNA transposons) (Schietgat et al, 2018). Retrotransposons can be further subclassified into four orders according to structural features and the life cycle of the element.…”