rsomics-seq-dist

Pairwise nucleotide-sequence distances from an aligned FASTA, value-exact to scikit-bio. Computes Hamming (proportion), Jukes-Cantor (JC69), and Kimura-2-Parameter (K2P) and emits a symmetric distance-matrix TSV that pipes directly into rsomics-nj-tree / rsomics-upgma.

Usage

rsomics-seq-dist aligned.fasta --metric k2p
rsomics-seq-dist aligned.fasta --metric jc69 -o dist.tsv
cat aligned.fasta | rsomics-seq-dist - --metric hamming

--metric is one of hamming (default), jc69, k2p. Input is an aligned FASTA (all records the same length) on a path or stdin (-).

Output

A header row of tab-separated ids, then one row per id with n tab-separated numeric cells — a full symmetric square matrix, the layout rsomics-nj-tree and rsomics-upgma read. Over-saturated or empty pairs are written as nan. Values are byte-identical to scikit-bio's DistanceMatrix.

Metrics

• Hamming — proportion of differing aligned positions over the full alignment length; gaps and ambiguity codes are compared literally (no exclusion), matching skbio.sequence.distance.hamming(proportion=True).
• JC69 — D = -3/4·ln(1 - 4/3·p) over the canonical p-distance. Non-canonical columns (gaps, N, ambiguity codes) are excluded pairwise. p ≥ 3/4 is over-saturated and returns nan.
• K2P — D = -1/2·ln(1 - 2P - Q) - 1/4·ln(1 - 2Q), with P/Q the transition/transversion proportions over canonical sites. A non-positive log argument (over-saturation) returns nan.

Performance

Sequences are pre-encoded to canonical codes once, then the JC69/K2P pairwise tally runs as a portable-SIMD loop (wide::u8x16, stable Rust) — lane-wise canonical-site, transition, and transversion masks reduced with move_mask + count_ones. Work is parallelised over matrix rows with rayon. Single-threaded, per metric, this beats scikit-bio's numpy-vectorized core on the same machine; multi-threaded and full-program it wins by a wide margin.

Origin

This crate reproduces scikit-bio 0.7.2's skbio.sequence.distance.{hamming, jc69, k2p}, whose source is BSD-3-Clause and was read directly. The evolutionary models are:

• Jukes & Cantor (1969), Evolution of protein molecules, Mammalian Protein Metabolism 3(21):132.
• Kimura, M. (1980), A simple method for estimating evolutionary rates of base substitutions through comparative studies of nucleotide sequences, Journal of Molecular Evolution 16(2):111-120.

Goldens are generated from scikit-bio 0.7.2 and committed under tests/golden/; the compat test runs without scikit-bio installed.

License: MIT OR Apache-2.0. Upstream credit: scikit-bio (BSD-3-Clause).