Onika is an efficient Rust MinHash sketcher and similarity engine built around compressed inverted indexes.

Please cite: Onika preprint – https://www.biorxiv.org/content/10.1101/2025.11.21.689685v1

git clone https://github.com/Malfoy/Onika.git
cd Onika
cargo build --release
# Binary is at target/release/Onika
target/release/Onika --help

• Single FASTA/Q: --input-fasta some_reads.fq[.gz|.zst] treats each record as one document.
• File-of-files (FOF): --input-fof samples.fof where samples.fof lists one FASTA/Q path per line. The order defines the genome IDs used in outputs (0-indexed).
• Both modes accept gzipped or zstd-compressed input transparently.

1. Inspect the bundled sample FOF (two genomes in data/):

   cat data/fof.txt

2. Build a sketch index from the sample FOF:

   ./target/release/Onika sketch --input-fof data/fof.txt --k_size 31 --s_size 10 --w_size 16  --reorder-similarity -o index.bin

3. Compare the built index against itself:

    ./target/release/Onika compare --ref-sketch index.bin --query-sketch  index.bin -o out.tsv.zst

   By default the output is Zstandard-compressed. Use --zstd-level 0 to write plain text.

Common

• --threads <INT>: override the auto-detected thread count.
• --zstd-level <LEVEL>: compression level for the index (sketch) or comparison output; 0 disables compression.

sketch

• --input-fof <FILE> | --input-fasta <FILE>: input source (one required).

• -o, --output <FILE>: where to write the sketch index (.bin is typical).

• -k, --k-size <INT>: k-mer size (default 31).

• -s, --s-size <INT>: log2 of sketch partitions; sketch has 2^s buckets (default 10 → 1024 partitions).

• -w, --w-size <INT>: fingerprint width in bits; values are reduced into a 2^w range (default 16).

• --reorder-similarity: greedily reorder sketches by self-similarity.

compare

• Reference inputs: --ref-sketch <FILE> | --ref-fof <FILE> | --ref-fasta <FILE>.
• Query inputs: --query-sketch <FILE> | --query-fof <FILE> | --query-fasta <FILE>.
• -o, --output <FILE>: destination for similarity results (Zstandard-compressed unless --zstd-level 0).
• --threshold <FLOAT>: similarity cutoff. Below this value hits are pruned in sparse mode; also controls how many bucket matches are required.
• --matrix: emit a full similarity matrix row per query (comma-separated floats). Without it, output is sparse.
• --prob-threshold-probability <FLOAT>: probabilistic pruning tail probability; set to 0 to disable the heuristic (default 0.001).

• Index (.bin): binary file containing the sketch table and metadata; compressed per-position blocks with stream-vbyte deltas and optional Zstandard. Load with --ref-sketch or --query-sketch.
• Sparse comparison (default): each line is query_<qid>\tref_gid:score[,ref_gid:score...], sorted by score descending. IDs are zero-based in the order the inputs were read. score is the fraction of matching buckets (matches / sketch_size).
• Matrix mode: each line is query_<qid>\t followed by comma-separated similarity scores for every reference (values below the threshold are written as 0.0000).

• Larger s-size adds resolution but increases memory and runtime; w-size controls fingerprint collision rate in the index.
• --reorder-similarity can improve spatial locality of sketches before compression; leave it off for the fastest build.
• Outputs and indexes are Zstandard-compressed by default; set --zstd-level 0 if you need plain text for downstream tools.

This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). See LICENSE for details.