amazon-science/uniqsketch v1.4.0

v1.4.0 — comparesketch --low-mem streaming mode

Repository: amazon-science/uniqsketch

Tag: v1.4.0

Published: 2026-07-16T18:50:14Z

Prerelease: no

Release notes:

Features

• `comparesketch --low-mem`: streaming M×N comparison for large reference sets.

The default mode pre-caches every reference's k-mer hashes (M+N file reads), but its memory scales with the entire reference universe — for a few thousand multi-megabase genomes this can exceed 1 TB. With --low-mem, only one Bloom filter and one reference's k-mers are held per thread at a time: each first-set reference is loaded into a filter, and every second-set reference is re-read from disk and queried against it. Memory becomes O(threads × genome) instead of O(all genomes), at the cost of M×N (rather than M+N) file reads. The similarity output is identical to the default mode; only the memory/I/O profile changes.

Full changelog: https://github.com/amazon-science/uniqsketch/blob/main/CHANGELOG.md