PanTE: A Comprehensive Framework for Transposable Element Discovery in Graph-based Pangenomes

Transposable element (TE) annotation is crucial for understanding genetics, genomics and evolution, yet current methods struggle to identify TEs in graph-based pangenomes. We developed a framework PanTE to construct accurate and representative TE libraries for both single genomes and graph pangenomes. PanTE is the first of its kind capable of being directly applied to graph-based pangenomes to build population-level TE libraries. By partially reimplementing RepeatModeler2 and integrating key innovations, including graph pangenome disassembly, alignment-free LTR structure detection, a machine learning-based classifier and efficiency-boosting strategies, PanTE outperformed RepeatModeler2 by efficiently handling large genomes, detecting high-abundance TEs and LTR-retrotransposons, and providing robust TE classification with superior computational efficiency. Compared to EDTA, it annotated ~ 26% more TEs in the grapevine genome and achieved up to 13 times faster runtimes in the wheat genome. PanTE represents a significant advancement in population-wide TE discovery, making it particularly valuable for pangenomic studies.