LTR_Retriever: A Highly Sensitive and Accurate Program for Identification of LTR Retrotransposons

Long-terminal repeats retrotransposons (LTR-RTs) are prevalent in most plant genomes. Identification of LTRs is important for studying the temporal change of genome size and facilitating the de novo gene annotation. Identification of LTR regions through homology search is usually inefficient due to its species-specific feature. Benefit from the conserved structure of LTR-RTs, multiple highly sensitivities programs were published for de novo LTR identification. However, false positives could dominate the predictions and excessive curations are usually required. Here we report LTR-retriever, a multithreading empowered Perl program that could process LTR candidates and generate high-quality LTR libraries regardless of input or genome quality. Benchmark tests on model plant genomes indicate that LTR-retriever has achieved high levels of sensitivity, specificity, accuracy, and precision, which are 92.0%, 96.8%, 95.7% and 89.88% in rice, respectively, demonstrating significant improvements compared to existing tools. We also explored the feasibility of identifying intact LTR-RTs from self-corrected PacBio reads. The LTR library constructed from 40 thousand reads covering 4.5X of the Arabidopsis Ler-0 genome surpasses that constructed from the genome. Our package has demonstrated the highest performance with great flexibility for retrieving LTR-RT.