EpiATLAS — Experiment Metadata Collation

This repository collects, harmonizes and enriches experiment metadata for EpiATLAS. It combines multiple provenance sources (EpiRR, ENCODE, local DEEP TSVs and local Excel sheets) to produce a single, analysis-ready experiment metadata table used by downstream pipelines.

Quick summary

• Purpose: Resolve and merge antibody & experiment metadata from ENCODE, EpiRR and local sources.
• Primary outputs: EpiATLAS_experiment_metadata_extended.csv (enriched metadata) and unfiltered long data experiment_metadata_extended_long.csv, which contains all information that was gathered in this effort (not only EpiATLAS reprocessed data).
• Primary entry point: collate_experiment_metadata.ipynb (notebook) — main workflow implemented as a Jupyter notebook.
• utils.py — helper functions for HTTP requests, ENCODE/EpiRR lookups, DEEP TSV parsing, flattening XML/JSON, and dataframe utilities.
• EpiATLAS_experiment_metadata.csv — input table from SFTP that drives lookups (place your updated input here before running).
• environment.yml — conda/micromamba environment specification for reproducible installs.

Additional Inputs

• Local DEEP TSV files under deep_experiments/ following the repository's expected naming conventions.
• Optional local Excel sheets (e.g., CREST exports) for additional antibody provenance.

Outputs

• EpiATLAS_experiment_metadata_extended.csv — enriched, merged experiment metadata (primary deliverable).
• experiment_metadata_extended_long.csv, which contains all information that was gathered in this effort (not only EpiATLAS reprocessed data) unfiltered and uncleaned.
• encode_roadmap_antibody_dicts.json — cached per-experiment antibody extracts from the encode portal.
• epirr_experiment_data.json — cached EpiRR experiment responses.

How it works (high level)

1. Load the input EpiATLAS_experiment_metadata.csv into a Polars DataFrame and validate basic assumptions (no entirely-null columns, expected key columns present).
2. Download the IHEC metadata spec from GitHub and materialize a mapping of expected columns per assay.
3. Query EpiRR for experiment details (caching responses to epirr_experiment_data.json) and normalize returned keys to match the EpiATLAS schema.
4. Discover and read local experiment metadata files (EGAX XML/JSON, DEEP TSVs, CREST Excel, EpiHK files, etc.), flattening nested JSON/XML into tables and adding a file/PRIMARY_ID provenance column.
5. Coerce and normalize assay and experiment type labels (e.g., map Bisulfite-Seq → WGBS, normalize RNA-Seq subtypes) so joins are consistent.
6. Merge EpiRR-derived records with local files using PRIMARY_ID and experiment_type where available; fall back to merging by epirr_id_without_version + assay_type when experiment_type is missing.
7. Enrich merged rows with provenance-specific lookups:

• fetch per-experiment antibody metadata from ENCODE / Roadmap (cached to encode_roadmap_antibody_dicts.json) and join on PRIMARY_ID;
• ingest local DEEP, CREST, EpiHK, GIS and manual CEMT metadata and coalesce into the master table.

8. Resolve duplicated and near-duplicate column names (case and minor spelling variants) — group and coalesce known groups and use fuzzy matching (Levenshtein) to find candidates for manual review.
9. Coalesce column groups so each logical metadata field has a single canonical column; enforce that at most one non-null value exists per canonical group per row.
10. Collapse and aggregate enrichment columns into list-valued fields where multiple provenance sources supply different values; drop placeholder/null tokens and normalize empty lists to null.
11. Validate that the enriched dataset still maps 1:1 to the input EpiATLAS experiments (no accidental loss or uncontrolled duplication), compute per-column completeness statistics, and produce diagnostic plots (heatmaps of completeness by centre/assay).
12. Export the final deliverables:

• experiment_metadata_extended_long.csv, which contains all information that was gathered in this effort (not only EpiATLAS reprocessed data) unfiltered and uncleaned.
• EpiATLAS_experiment_metadata_extended.csv (enriched, merged metadata),

Notes:

• The notebook is iterative — cached JSONs and locally persisted intermediate CSVs speed repeated development.
• Any new provenance source should add a loader in utils.py, a small mapping into the merge logic, and accompanying tests for uniqueness/coalescing.

Notes on caches and development

• The first runs will query external APIs (ENCODE, EpiRR). To speed iterated development, the notebook creates and reuses epirr_experiment_data.json and encode_roadmap_antibody_dicts.json.
• utils.py contains the HTTP retry logic and the TSV / JSON/XML flattening utilities. If you add new provenance sources, extend utils.py and the notebook mapping tables.

Contact

open an issue

---

(README created with generative AI and corrected by Quirin Manz.)