As long-read RNA-seq datasets scale from single experiments to population-level cohorts, one practical problem becomes increasingly important:

How can we efficiently merge and compare millions of transcripts across hundreds or thousands of GTF files?

I am sharing isomatch, a Rust-based toolkit for large-scale GTF processing, with a focus on long-read RNA-seq transcript merging, structural classification, and provenance tracking.

In short, isomatch:

• performs wobble-tolerant splice-junction matching, with separately configurable handling of canonical splice sites (GT-AG / GC-AG / AT-AC) and non-canonical sites;

• supports TSS/TES-aware merging, since transcripts with the same intron chain are not always equivalent isoforms;

• supports guide-based representative selection using resources such as refTSS and PolyASite;

• provides SQANTI3-compatible structural classification for large transcript sets;

• preserves full provenance, so each merged transcript can be traced back to its original source transcript and sample.

The tool is designed for population-scale long-read RNA-seq datasets. In our large-scale tests, merging across thousands of samples used around ~10 GB memory.

It is distributed as a single static binary with no external dependencies.

GitHub: https://github.com/zhengxinchang/isomatch

Feedback, suggestions, and test cases from people working with large long-read RNA-seq transcriptome datasets would be very welcome.