SEPIA

Systematic Evaluation of Pigmentation In Archaic hominins

SEPIA asks what skin/hair/eye pigmentation genetics can — and can't — tell us about Neanderthals and Denisovans, and how they compare to modern humans. It is a careful, QC-first re-analysis: the archaic DNA is patchy and shallow and pigmentation genes don't behave like ancestry markers, so the contribution is an honest map of what the pigmentation genetics does and doesn't support, with the uncertainty made explicit.

This repository is the working anchor for the project. It continues and re-evaluates the prior honors thesis PAINT (Lily Heald), rebuilding the analysis from clearly-traced inputs; Lily's original thesis repository is separate and unchanged.

Where things are

• plan/plan.md — the single canonical roadmap (phased; bug fixes, analysis redesign, QC checks Q1–Q10, infrastructure, writing). Shareable rendering: plan/sepia-plan.html.
• plan/changelog.md — dated log of what was done / found / decided.
• plan/docs.html — hub linking every project document (data & pipeline record, rebuild plan, per-genome provenance & QC, bug-evidence log).
• data/README.md — per-file data provenance, and where the large / cluster-only data live.

Layout

The website and the compute pipeline live in one repo; on the cluster the repo is the working directory, so results flow into git while big data stays out of it.

*.qmd  _quarto.yml   Quarto website (source + config) → GitHub Pages
analysis/            site helper scripts + the introgression Shiny app source
code/                cluster pipeline (SLURM/bash): merge, QC, PCA
plan/                roadmap, changelog, provenance & QC docs (+ rendered HTML)
data/                small tracked inputs (see data/README.md)
output/              small tracked RESULTS that sync (PCA coords, QC tables, figures)
resources/  scratch/ gitignored, cluster-only: genomes + big intermediates (never synced)

On Great Lakes the checkout lives at /nfs/turbo/lsa-tlasisi1/sepia; reusable reference genomes live in a shared …/genomes dir (referenced via $SEPIA_GENOMES or the gitignored resources/genomes symlink) so other lab studies can call the same data without copies.

Build

• Website (Quarto): quarto render. R-heavy analysis pages use freeze: auto — they render from committed _freeze/ in CI (which does not run R), so they must be rendered once locally/on the cluster (with R + the data) to populate _freeze/. Prose pages render anywhere. .github/workflows/publish-site.yml deploys the rendered _site/ to GitHub Pages (currently set to manual workflow_dispatch until the first full render).
• Project docs under plan/: regenerate the HTML with bash plan/build_docs.sh.
• Pipeline: run code/*.slurm from the cluster checkout (sbatch code/…).

Team

Tina Lasisi (PI / advisor) · Lily Heald · Yemko Pryor. Code under the MIT License; written content and figures under CC BY 4.0.