IsoMatch enables population-scale transcript integration and reveals reference-dependent isoform novelty

Abstract

Transcriptome analyses increasingly generate large collections of transcript models that must be integrated, compared and interpreted across samples and reference annotations. Transcript novelty depends on structural equivalence criteria and reference annotation. Here we present IsoMatch, a scalable framework that harmonizes transcript merging and reference-based classification through explicit, configurable structural criteria. IsoMatch preserves source-GTF and original-transcript traceability throughout the workflow, allowing every merged and classified model to be traced back to its contributing inputs. Benchmarking showed that IsoMatch preserves distinct transcript models during integration through configurable matching parameters for both splice junction chains and transcript ends. With its index-based architecture, IsoMatch merged 163 ENCODE transcript catalogs and classified 7.9 million transcript models with peak memory usage of 0.49 GB and 1.1 GB, respectively. Integrating GENCODE, RefSeq and CHESS revealed that 71.2% of transcript groups remained annotation-specific even without terminal constraints. For 4.84 million merged transcript models from 119 long-read transcriptomes from the 1000 Genomes Project, the integrated reference increased the full-splice match proportion from 12.0-13.6% with individual annotations to 18.0%. IsoMatch enables scalable transcript integration and classification with explicit structural criteria and preserved traceability, demonstrating how reference choice shapes estimates of transcript novelty.