Batch, Prefetch, Skip, Orient, Squeeze: Faster, Smaller k-mer dictionaries via the Spectral Burrows-Wheeler transform

A batched, prefetching SBWT k-mer index with orientation squeezing achieves up to 4.9× faster streaming lookups than SSHash for 31-mers, using 12 bits per canonical k-mer.

• Why it matters: Streaming k-mer queries are central to read mapping and metagenomic classification, but SBWT indexes lag behind hashing-based SSHash in speed despite being space-competitive.
• What they did: The authors batch streaming queries with prefetching and a rank structure that reads one cache line per query, extend batching to prune many k-mers after a failed lookup, and store only one k-mer orientation.
• The result: Across SSHash benchmarks and two larger collections, their fastest SBWT answers streaming 31-mer queries faster than SSHash on every read set (up to 4.9×), while using 12 bits per canonical k-mer versus SSHash’s 7.4–11.1.

AI summary of the abstract · may contain errors · not medical advice · read the paper.