A Transformers Analysis Pipeline (TrAP) to Evaluate LINE-1 RNA Sequence Content

LINE-1 RNA has emerged as a potential regulator of chromatin organization and gene expression, as well as a candidate readout of cellular and disease-associated states. Accurate measurement of LINE-1 transcript abundance is therefore important for evaluating its regulatory, biomarker, and potential therapeutic relevance. Measurement is challenging in short-read RNA sequencing because similar LINE-1 loci produce ambiguous, multi-mapping reads and LINE-1-associated signal can arise from passive as well as autonomous transcription. We developed the Transformers Analysis Pipeline (TrAP), an alignment-free ALBERT-based workflow that classifies short RNA-seq reads as L1HS, L1PA, or non-LINE-1 before abundance estimation, converting overlapping k-mers to integer tokens and integrating predictions with Salmon- or expectation–maximization-based quantification. TrAP was evaluated using synthetic LINE-1 insertion datasets and HEK293 short-read RNA-seq benchmarked against 5′ rapid amplification of cDNA end (5′RACE) long-read measurements. AlbertSalmon agreed most closely with the synthetic insertion levels, ranking first in all insertion levels and deletion-probability strata. Agreement with the 5′RACE reference was also highest, although the locus-level gain there was only incremental. High-attribution k-mers were modestly over-represented within functional ORF2 coding regions rather than being uniformly distributed across LINE-1. These findings support TrAP as a transformer-assisted framework for improving LINE-1 quantification in short-read RNA-seq datasets, thereby supporting future regulatory and biomarker studies.