Abstract We introduce branchwater, a flexible and fast petabase-scale search for the 767,000 public metagenomes presently in the NCBI Sequence Read Archive. Our search is based on the FracMinHash k-mer sketching technique and can search all public metagenomes with 1000 query genomes in approximately 36 hours using 50 GB of RAM and 32 threads. Branchwater is a Rust-based multithreading front-end built on top of the sourmash library. We provide biological use cases, examine performance, and discuss design and performance considerations.

Sourmash Branchwater Enables Lightweight Petabyte-Scale Sequence Search