Transcript quantification using context-aware MAPQ modeling
Abstract
Long-read RNA-sequencing enables transcript-level quantification, but filtering alignments using fixed mapping quality (MAPQ) thresholds can discard reads reflecting intrinsic transcriptome ambiguity and distort abundance estimates. Here we present CALMeR (Context-Aware Long-read MAPQ-based Rescuer), a regression-based approach that fits transcript and read features. By modeling the expected mapping quality of each alignment, the method distinguishes low scores consistent with transcript context from anomalously low scores indicative of artifacts. This approach retains reads from multi-isoform genes and paralog families that would otherwise be excluded, while removing alignments with unexpectedly low confidence. Retained reads can be counted per transcript or supplied to downstream quantification tools. In spike-in benchmarks, context-aware filtering recovers more transcripts and yields abundance estimates that better match known concentrations than a MAPQ ≥ 1 filter and expectation-maximization-based quantification methods. These results support incorporating transcript context into alignment filtering for long-read RNA sequencing. Open-source software is available at https://github.com/sparthib/CALMeR.