mtnoc-rs

mtNoC — a de-identified estimate of the number of distinct human maternal (mtDNA) contributors to a shotgun-metagenomic sample. It reports a conservative lower bound on the count of distinct maternal lineages (macro-clades) plus their composition — aggregate, never an individual haplotype.

Pure-Rust reimplementation of the reference Python pipeline. A single static binary takes raw FASTQ in and emits one TSV row per sample.

Pipeline

raw FASTQ.gz ─▶ k-mer prefilter (rCRS 21-mers) ─▶ align to chrM ─▶ per-read alleles at
               PhyloTree-informative positions ─▶ mixture-EM + bootstrap-stability ─▶ contributor
               lower bound (macro/sub-clade) + 95% CI + lineage composition

Two modes:

• fast (100% pure Rust): raw FASTQ pair → embedded-rCRS k-mer prefilter → a triple_accel SIMD seed-anchored aligner (against the compact reference) → per-read alleles. No external tools.
• accurate: raw FASTQ pair → strobealign → samtools filter (MAPQ≥20, NM/qlen≤0.027) + mpileup -Q20 allele extraction → reproduces the reference pipeline exactly. Pass a pre-aligned chrM BAM with --bam to skip strobealign. Shells out to strobealign + samtools.

Both modes feed the same mixture-EM, which is bit-for-bit identical to the reference Python implementation (including a numpy-compatible MT19937 so the bootstrap matches exactly).

Install / build

cargo build --release          # target/release/mtnoc
# or, once published:  cargo install mtnoc-rs

Reference data

The EM reference dump and the rCRS k-mer prefilter sequence are embedded in the binary (gzip, decompressed at startup), so you only supply the alignment reference:

• data/mtNoC_compact.fa — chrM (NC_012920.1) + NUMT-decoy contigs. Ships in the git repo; pass it via --ref. fast uses it for the pure-Rust aligner; accurate uses it as the strobealign index and mpileup reference.

data/mtref.txt (regenerable via scripts/dump_ref_for_rust.py) and data/rcrs_wrap.fa are the sources for the embedded tables; you do not need to pass them.

Usage

# FAST (pure Rust) — one sample
mtnoc fast --in1 S_R1.fastq.gz --in2 S_R2.fastq.gz --ref data/mtNoC_compact.fa

# Shorthand: give the sample stem, mtnoc resolves _R1/_R2 (or .R1/.R2) + extension
mtnoc fast --in S --ref data/mtNoC_compact.fa

# Multiple samples (one TSV row each): comma-separate, or comma-separate stems with --in
mtnoc fast --in1 A_R1.fq.gz,B_R1.fq.gz --in2 A_R2.fq.gz,B_R2.fq.gz --ref data/mtNoC_compact.fa
mtnoc fast --in A,B --ref data/mtNoC_compact.fa

# Optionally save the chrM-candidate reads (omit to only emit the estimate)
mtnoc fast --in1 S_R1.fastq.gz --in2 S_R2.fastq.gz --ref data/mtNoC_compact.fa \
           --out1 S.cand_R1.fq --out2 S.cand_R2.fq

# ACCURATE (needs strobealign + samtools on PATH)
mtnoc accurate --in1 S_R1.fastq.gz --in2 S_R2.fastq.gz --ref data/mtNoC_compact.fa
# ...or from a pre-aligned chrM BAM (skips strobealign)
mtnoc accurate --bam S.chrM.bam --ref data/mtNoC_compact.fa

Common options: --thread N (0 = all cores), --header (prepend a column-name line), --n-boot, --seed, --min-prop, --min-identity (fast). Progress goes to stderr; only the TSV goes to stdout.

Output

Tab-separated, one row per sample, no header unless --header is given (so multiple runs concatenate cleanly). Columns:

Example:

sample           informative_reads   contributors_ci95   composition
mix_EAS85_AFR15  4000                1;1;3               M7c2b:0.9770:3736r;Z2:0.0230:88r

min_contributors_subclade is a floor limited by the bootstrap-stability criterion (≥8 supporting reads, stable in ≥80% of resamples); the true contributor count is generally higher.

Dependencies

• Build: Rust ≥ 1.74. All crates are pure Rust — no C toolchain, no *-sys build steps. gzip is flate2 (zlib-rs backend); FASTX parsing is needletail (default-features = false).
• Runtime:
  • fast — none. A single self-contained binary.
  • accurate — external strobealign and samtools on PATH (override with --strobealign / --samtools). Only this mode uses external tools.

Ethics

Reports only aggregate counts and maternal-lineage composition — never an individual's haplotype or a re-identifiable profile.

Attribution

Haplogroup definitions derive from PhyloTree (van Oven & Kayser); rCRS = NC_012920.1. Aligner kernels: triple_accel. Please cite PhyloTree and this tool if used.

License

MIT — see LICENSE. (Bundled reference data derived from PhyloTree/rCRS retains its original terms.)