Compact by design
Only the fields a quantifier needs, at the integer widths that fit them. No sequence, no quality strings, no SAM boilerplate.
RAD format
Reduced Alignment Data — a compact, parallel-friendly, self-describing binary record of how sequencing reads map to a set of targets.
What is RAD?
Section titled “What is RAD?”RAD (Reduced Alignment Data) is a binary format for recording how sequencing reads map to a set of targets — a transcriptome, genome, or metagenome. It is reduced because it is free to carry less information than a SAM/BAM file: only what a downstream quantification tool actually needs. In exchange it is compact and cheap to parse in parallel, which is the whole point.
RAD files are produced by piscem and
salmon, and consumed by
alevin-fry and
piscem-infer. The reference
reader/writer is libradicl.
Built for parallelism
A file is a stream of independent chunks. One thread reads chunks off disk and hands them to a pool of workers — no per-record locking.
Self-describing
A tag system declares the dynamic fields at each level; a reader learns the record layout from the file, not from a hard-coded assumption.
Versioned & extensible
A magic signature plus a spec version lets the format grow back-compatibly — new optional fields and semantic roles are additive minor bumps.
Where to start
Section titled “Where to start”- New to the format? Start with What is RAD? and the motivation & design principles.
- Implementing a reader or writer? The file structure page is the top of the byte-level specification.
- Producing v2 files? See spec versioning and tag roles.