The architecture of pigmentation genetics
One coupled pigmentation–hormone gene network, pointed at two questions — human association grading, and the evolution of primate sexual dichromatism
Sexual dichromatism — males and females of a species differing in color — arose ~15 times independently across the primate radiation and is polygenic in every origin tested, behaving like a coupled pigmentation–sex-hormone system with several accessible routes rather than a single switch (MC1R is not the hit). This is the first clade-wide, multi-gene selection test of the genomic basis of primate sexual dichromatism. (For contrast, the best-studied bird cases rest on a gene of major effect — MC1R/melanin in galliforms, Nadeau et al. 2007; BCO2/carotenoid in finches, Gazda et al. 2020 — a useful cross-taxon comparison, not the main result.) Whether the architecture is shared or heterogeneous across origins is currently underpowered to resolve (see the finding page). → Read the finding
This project builds one thing — a curated, coupled pigmentation gene network (with its physically-linked sex-hormone axis) — and points it at two questions. Pointed at human genetics, the network grades how much to trust a claimed variant→phenotype link by making independent evidence layers converge. Pointed at evolution, the same network yields a genuine discovery: primate sexual dichromatism is not one gene but a polygenic, coupled pigmentation–hormone trait — which is why it arose ~15 times independently and is so readily lost. Same object, same one-gene-is-not-enough lesson, two scales.
Our integration, not any single source. The harmonized substrate in three dimensions — four independent evidence layers, one gene core, with each gene sitting only on the layers that carry it. Drag to rotate · hover a gene.
Short on time? Read the dichromatism finding first — it is the discovery: sexual dichromatism arose ~15 times independently across primates and is polygenic in every origin tested (unlike birds, where dichromatism maps to genes of major effect like MC1R and BCO2), tested with per-origin RELAX, per-branch aBSREL, and RERconverge.
Then, for the human-pigmentation method, follow Reconstruct the network → Unified association base → Mechanism→direction law → Direction law, expanded. The direction-law notebook shows a functional screen ordering the direction of Mendelian pigmentation disorders (22/22, permutation p < 1×10⁻⁵); read it with its expansion, which extends the result to 29/33 and locates its failure boundary. The human arm is a methodological demonstration of the convergence thesis on well-characterized pigmentation biology. Every code cell is folded — click “Show code” to expand it.
What this is
One curated pigmentation gene network — including its physically-coupled sex-hormone axis — applied to two questions that make the same point: the map from genotype to pigmentation is not one-gene-one-trait; you have to read the genes as a coupled network.
Across primates (the discovery). Sexual dichromatism — a trait that maps to genes of major effect in birds (MC1R, BCO2) — turns out to have arisen ~15 times independently in primates and to be polygenic in every origin tested, drawing on the coupled pigmentation × sex-hormone network rather than MC1R. Whether different origins recruit the same or different parts of that network is the open, currently-underpowered question. See the finding →
Within humans (the method). To know how much to trust a claimed variant→phenotype link, make independent lines of evidence converge — statistical association, mechanistic pathway knowledge, and experimental validation. Applied to human pigmentation, this grades reported associations and names the next step that would confirm each; it shows the same variant can be causal in one population and silent in another (context = ancestry), and that the direction of an effect is predictable from mechanism.
Pigmentation is the model system because it is highly heritable, has near-zero environmental variance, has a well-characterized molecular mechanism, and shows these context effects in a visible, checkable form — from a single human variant read across ancestries to a whole trait reassembled across a primate radiation.
Everything here — every notebook, table, and figure — was built during the hackathon event window (first commit 2026-07-09). Each notebook is a self-contained mini-manuscript (intro → methods → results → discussion, with citations) that renders from its own stored outputs, so the site needs no kernel, no network, and no withheld data. Code is folded by default — click “Show code” on any cell to see it.
One network, two questions
- Pointed at evolution — sexual dichromatism across primates (the discovery): using the coupled pigmentation–hormone network to test selection across ~15 independent origins, establishing the trait is polygenic (no single gene of major effect as in birds) and labile, while treating shared-vs-heterogeneous architecture as an open, power-limited question. Pipeline and cluster protocol in
comparative-genomics/. - Pointed at human genetics — the convergence method (the notebooks below): using the same network to grade how much to trust human pigmentation genotype→phenotype claims by making independent evidence layers converge.
The build, in three parts
1 · Building the pigmentation network
- Reconstruct the network — rebuilds the Raghunath et al. 2015 melanogenesis model as a signed, directed network, typing and signing every interaction from the published file, with a resolvable citation per edge.
- Resolve to genes — resolves entities to genes (UniProt / HGNC / Ensembl / GRCh38) and validates every edge against OmniPath as a separate check; a closing gate fails the build if anything is uncited.
- Validation cases — pulls published genotype→phenotype discordance cases from curated papers, counted paper-by-paper, reproduced from committed extracts.
- Unified association base — a claim-faithful effector-status re-read of 105 curated loci: 14 genuinely effector-uncertain (34 with Kim 2024 folded in) vs 75 canonical-effector variant-gaps — correcting an earlier coarse “author-unexplained” tag that overstated the open set ~6×.
- Compare candidate networks — “the network you choose changes the answer”: how STRING drift and source choice move which genes are in play.
- Gene-regulatory network — 58 curated, signed TF→target regulon edges (MITF / PAX3 / SOX10), every edge cited.
- Harmonized substrate — merges all layers into one tagged node/edge table with per-edge provenance and tier.
2 · Adding the hormone axis
- Sex-hormone layer — layers the sex-hormone axis onto the pigmentation substrate through cited hormone → melanocyte-receptor → MITF bridges, each gated on melanocyte expression (Zhang eQTL). The estrogen bridge runs through membrane GPER (not ERα — melanocytes are ERα-negative), converging on the same cAMP→CREB→MITF node as melanocortin. It is a separate, additive layer: the pigmentation network built above is unchanged.
- Gene panel — two coupled networks — assembles the combined pigmentation + sex-hormone gene panel used for the cross-species work, with each gene’s ortholog-mapping confidence flagged so comparisons rest on clean one-to-one orthologs.
3 · Evolutionary questions
- Dichromatism synthesis — brings the evolutionary results together: how many times primate sexual dichromatism arose independently, whether its genetic architecture is polygenic rather than a single gene of major effect, and how labile the trait is across the tree.
Sidequests
Self-contained detours, kept out of the main build. Each is a question we followed up on and answered on its own terms; none is required to follow the three parts above.
- Rescue-screen diagnostic — can a locus whose causal gene is unknown be rescued by linking it into the melanogenesis network through an independent line of evidence? No novel effector emerges; the durable result is that resolving eQTLs in melanocytes rather than bulk skin retracts false causal-gene calls — the tissue you resolve in changes the answer.
- Bajpai orphan reconciliation — a symmetric STRING seed reconnects 93 of 142 CRISPR “orphans” to the core (vs 0/142 under a curated-only seed) — seeding changes the answer; the residue is the knowledge gap.
- Mechanism → direction law — a functional melanin screen orders the direction of Mendelian pigmentation disorders: positive regulator + loss-of-function allele → hypopigmentation (22/22, base rate 54%, p < 1×10⁻⁵), robust to the study-bias confound.
- Cross-ancestry conditionality — some pigmentation genes are discovered through different, population-private variants in different ancestries, quantified with Hudson FST against a genome-wide baseline.
- Direction law, expanded — a pre-registered expansion (29/33) that also locates the failure boundary: the misses act through systemic routes.
Appendix — data extraction
Reproducible pulls of each external dataset, frozen for provenance: Bajpai CRISPR screen · Baxter gene list · HIrisPlex markers · GWAS Catalog pull.
Reproducibility
Every figure and number renders from committed outputs; external pulls are frozen under data/external/db_responses/ with provenance sidecars. The authoritative build log is internal/CHANGELOG.md; the full repository is on GitHub.