dnasp

Population genetics of pre-aligned DNA sequences or multi-sample VCFs using selected DnaSP 6 methods. Use for diversity, neutrality statistics, linkage disequilibrium, InDel polymorphism, divergence, MK, Ka/Ks and codon usage; not alignment, phasing or clinical interpretation.

Trigger

Fire when a user requests population-genetic analysis of aligned DNA, a supported VCF, or a DnaSP-compatible statistic listed below. Do NOT fire for sequence alignment, read mapping, haplotype phasing, clinical advice or unsupported coalescent significance tests.

Scope

Analyse genetic variation in supplied alignments using 16 selected DnaSP methods. This is not a complete replacement for the DnaSP GUI or all its analysis modes. Read the statistical reference for definitions, exclusions, source conventions, file formats, examples and release validation evidence.

Core Capabilities

• Polymorphism: S, Eta, haplotypes, Hd, VarHd, pi, k, theta-W, G+C, Tajima's D, Fu and Li D*/F*, Ramos-Onsins and Rozas R2.
• LD: D, D', R2, ZnS, Za, ZZ and original-column pair labels; recombination Rm.
• Mismatch distribution with unbiased variance over unordered pairs and Sokal & Rohlf corrected CV; Harpending (1994) raggedness; Model 1 diallelic InDel diversity.
• Divergence and Hudson Fst between populations; outgroup Fu and Li D/F.
• Two-locus HKA, McDonald-Kreitman, Nei-Gojobori Ka/Ks, Fu's Fs and SFS.
• Ts/Tv, codon counts/RSCU including stops, named per-sequence ENC and its synonymous-codon-weighted summary.
• Sliding windows mirror DnaSP's Gaps in Sliding Window = considered mode: starts advance from 1 by the step, capped at alignment length; ends are capped there too, and the first window reaching the end terminates the loop. Empty windows are retained. The not-considered mode is not implemented.
• Window midpoints are the alignment position of the ceil(net/2)-th gap-free column, or the window start if none exists; exported in TSV, JSON and report and used for the plot. VCF windows use retained SNP indices, including the final partial window.
• Raw per-site Fay-Wu H and theta-L minus theta-W; normalised Hn/ZE are absent.

Fu and Li D*/F* and outgroup D/F mirror Data > Segregating Sites/Mutations = Segregating sites, using DnaSP's v5-style panels. The rp49 Eta-setting D*/F* figures can be reproduced by substituting eta for S, but this is not a general conversion: FULI.vb also changes singleton and external-mutation capping (SingleMut and ExternaMut subtraction). No Eta-mode switch is implemented.

Workflow

1. Confirm the input is pre-aligned and identify the scientific comparison: ingroup, explicit outgroup, populations, coding interval and genetic code. Do not infer an outgroup from record order or silently select a code.
2. Run dnasp.py from this skill's own directory; it is self-contained (its reproducibility writers ship beside it as _repro_writers.py). Install matplotlib for figures.
3. Choose an empty output directory. For alignment input use --input; for VCF use --vcf; for two-locus HKA alone use --hka-file --analysis hka.
4. Select actual implemented names with --analysis. Supply --outgroup, --input2 or --pop-file when needed. Use --genetic-code vertebrate-mitochondrial for the matching mitochondrial table. The default is standard. Coding intervals must be preselected and divisible by three.
5. Execute the CLI. An explicit analysis that cannot run returns a non-zero code. --analysis all is opportunistic: inspect its completed/skipped manifest.
6. Read stderr diagnostics, report.md and reproducibility/manifest.json. Distinguish a failed analysis, an undefined statistic and an excluded site.
7. Interpret the chosen statistic within its documented assumptions. Do not turn a signed value or an uncalibrated threshold into a significance claim.
8. Retain the input archive, settings, hashes and environment with the report. commands.sh replays on the recorded host/code path into a new output folder; moving a run requires the code and dependencies as well as its input archive.
9. For Windows GUI comparison, follow the separate validation checklist and capture raw DnaSP output. Source-derived expectations are not GUI observations.

Example Output

bash

python <skill-dir>/dnasp.py --demo --output new_demo_run
python <skill-dir>/dnasp.py --input alignment.fas --analysis polymorphism,ld --output new_run
python <skill-dir>/dnasp.py --input coding.fas --outgroup OutSeq --genetic-code vertebrate-mitochondrial --analysis mk,kaks,codon --output new_coding_run
python <skill-dir>/dnasp.py --vcf samples.vcf --analysis polymorphism,sfs,fufs --output new_vcf_run

<skill-dir> is this skill's own directory: ${HERMES_SKILL_DIR} on Hermes, {baseDir} on OpenClaw.

The synthetic demo contains 10 ingroup sequences, one outgroup and 300 sites:

Quantity	Demo value
S / haplotypes	5 / 8
Hd / Tajima D	0.9556 / 0.6789
MK Pn / Ps / Dn / Ds	2 / 3 / 2 / 1
MK alpha	0.6667
Ka / Ks / omega	0.010239 / 0.030291 / 0.3380
Ts / Tv	4 / 1

The demo runs 15 modules; HKA uses a separate two-locus input. The regression suite, rather than the printed banner alone, checks the expected figures.

Output Structure

text

output/
  report.md
  results.tsv                  # polymorphism and windows
  summary.json                 # module summaries, window midpoints, named ENC/null
  result.json                  # structured envelope: headline summary, summary.json payload, artifacts
  ld_pairs.tsv                 # when LD pairs exist
  figures/                     # when matplotlib is available
  reproducibility/
    inputs/
    commands.sh
    environment.yml
    manifest.json
    checksums.sha256

Dependencies

Python 3.10 or later. Core estimators and the reproducibility bundle use the standard library. Plotting uses matplotlib; without it the statistics, report and TSV files are still written and figures are skipped. Install with pip install -r requirements.txt or use environment.yml. The Windows validation runner targets Python 3.12.

Gotchas

• The model will want to count IUPAC symbols as alleles. Do not. Supported ambiguity symbols are missing data and the nucleotide mask excludes their columns.
• The model will want to analyse an unknown outgroup as an ingroup-only run. Do not. Identifiers must be unique and an explicit outgroup must resolve.
• The model will want to call VCF diversity per-base diversity. Do not. VCF columns represent retained variant records; invariant callable bases are absent.
• The model will want to use any coding annotation in a NEXUS file. Do not. This implementation requires a preselected coding alignment and does not read CHARSET coding annotations. In COII, use positions 1-681 for codon usage.
• The model will want to apply one stop-codon rule everywhere. Do not. Selected stops count in RSCU and as family 21 in MK/KaKs; ENC and coding G+C use sense codons.
• The model will want to read raw H/E as normalised DnaSP Hn/ZE. Do not. These outputs are explicitly different, and no significance test is supplied.
• The model will want to overwrite an earlier output folder. Do not. Choose a fresh folder; runs reject non-empty destinations to preserve prior artefacts.
• The model will want to infer genomic-scale performance from the tiny bundled VCF examples. Do not. LD enumerates all biallelic-site pairs and uses quadratic storage; read the measured limits in the reference before a large run.

Safety

All sequence analysis and output remain local. This skill is a research and educational tool. It is not a medical device and does not provide clinical diagnoses. Consult a healthcare professional before making any medical decisions.

Agent Boundary

The agent selects documented inputs/options, executes the skill and explains reported results. The code computes the statistics. Neither the agent nor the skill may fabricate GUI validation, P-values or missing results.

Integration

Call dnasp.py directly. Use --analysis for module selection, not invented flags such as --pi, --kaks or --n-sim. The implemented options include VCF input, populations and genetic-code selection; --help lists them all.

Chaining Partners

Use an alignment tool before this skill when sequences are not aligned. A VCF filtering/phasing workflow may prepare input, but every filter and phase choice must be recorded. Downstream reporting may use the Markdown report and available TSV files; the TSV does not contain every module's results.

Maintenance

Recheck source/help-derived regression cases and the 170 historical comparison fixtures after changes to formulas, masks or parsers. Review GUI differences when the target DnaSP build or analysis mode changes. Keep this file, the method reference, CLI metadata, version and catalogue consistent. New Windows GUI observations must be reviewed before changing published concordance counts.