Automated Construction of High-Quality LTR Exemplars from Plant Genomic Sequences

Long terminal repeat (LTR) retrotransposons represent the largest component in most plant genomes. Construction of the LTR library becomes one of the critical tasks in a genome annotation project, as well as understanding the evolution of genome size across species. Ab initio discover of LTR elements computationally has been achieved based on their highly structured characteristics, however extensive efforts are still required for library constructions, often due to false positives and significant redundancy. We observed that most false positives generated by ab initio LTR searching software are due to the vicinity of highly similar repetitive elements but LTR retrotransposons (e.g., LINE, SINE, solo-LTRs, DNA transposons, tandem repeats, and etc.). By taking advantage of these features, we developed a package that distinguishes true and false LTR elements. Moreover, we reduced the redundancy to minimize the computational requirement for genome annotation. When testing in the rice genome (Oryza sativa c.v. Nipponbare) which has a well curated TE library, the constructed LTR exemplars had 94.7% specificity and 94.6% sensitivity. Among all LTR-searching software, LTR-modeler has shown the highest overall performance. With tests in various plant species, e.g., the maize genomes (Zea mays c.v. B73 and Mo17), the Arabidopsis genome and the sacred lotus genome (Nelumbo nucifera) , our package has shown well adaptable and accurate to construct LTR exemplars in plant genomes.