HDMA

AGENTS.md — HDMA-public

Orientation for AI coding agents working in this repository. (Claude Code users get the same guidance automatically via the shipped hdma skill — see .claude/skills/hdma/.)

What this repo is

The public code + data-access companion for the Human Development Multiomic Atlas (HDMA), a single-cell multiome (RNA + ATAC) atlas of human fetal tissues with per-cell-type ChromBPNet models, a motif lexicon, and enhancer/variant analyses. It is a reproducibility companion, not a turnkey pipeline. A fresh clone contains code + rendered HTMLs only; all data lives on Zenodo. Genome build throughout is hg38 / GENCODE release 42; analyses assume a Slurm HPC.

Cross-cutting concepts

• Data is on Zenodo, not in the repo. Pull it from https://zenodo.org/communities/hdma; outputs/data are gitignored.
• Cluster naming has two schemes: cluster_id / RNA_Clusters (Table S2, “c”-prefixed) vs. Cluster_ChromBPNet (Table S2 col 19, e.g. Adrenal_c0). The latter names almost all ChromBPNet/motif/bigwig/training-region files.
• Supplementary tables are the keys: Table S14 (data type → URL/DOI/record ID), Table S2 (cell metadata / cluster map), Table S6 (motif lexicon). All three ship in tables/ (table_s14.tsv, table_s2.tsv, table_s6.tsv; column dictionaries in tables/README.md).
• The repo’s own docs are authoritative: DATA.md/DATA.html, README.md, code/03-chrombpnet/README.md, MOTIFS.html.

Where to go

If the task is about…	Read
data — download/find/load, formats, browser tracks, SRA reprocessing	.claude/skills/hdma/references/data.md, then DATA.md
code — run/adapt a step, envs, figure→code, repo layout, path config	.claude/skills/hdma/references/code.md, then README.md

Three footguns to know up front

1. Data isn’t in the repo — download from Zenodo via tables/table_s14.tsv + the get_urls.py helper (see the data reference and DATA.md).
2. Create code/ROOT_DIR.txt (gitignored) with the repo’s absolute path before running any script — scripts read it instead of hardcoding paths.
3. Cluster files are named by Cluster_ChromBPNet (e.g. Adrenal_c0), not cluster_id.

This site is open source. Improve this page.