Recognition of SINE 3'-Ends Throughout the Tree of Life

According to the latest researches, retrotransposons (SINE, LINE) were found in all eukaryotic organisms and occupied a significant part of the majority of the genomes (42% for human, 36% for mice and 78% for a maize genome). A distinctive feature of SINE is the absence of regions that encode enzymes responsible for retrotransposition. It was deduced that SINE retains the ability to spread across the genome due to the similarity of the stem-loop structure at the 3'-end with the same stem-loop of homologous LINE. According to this, we assumed that stem-loop is important structure and should have similar properties for different SINE classes. The purpose of this study is to build machine learning models for stem-loop recognition at the 3'-ends of various SINE classes (derived from tRNA, rRNA, snRNA, 7SL), and also to identify features that are shared by different SINE. The data for the research consist of 822 patterns of transposons of various organisms throughout the tree of life. In this research, models were developed for distinguishing real data from artificially generated and classifying real sequences by their origin. The features used to train models were the nucleotide composition and physicochemical properties of the dinucleotides that make up the stem. As a result, there were trained models distinguishing classes with an accuracy of ~ 74%. Feature importance analysis shows that the most important features are the minimum free energy of the secondary structure of RNA and parameters hydrophilicity, shift and slide of the dinucleotide at the base of the loop. The results suggest that the stem-loop is indeed a significant structure of SINE retrotransposons, which has common properties for different SINE classes and for different types throughout the tree of life.