#!/usr/bin/env python3 """Conservative ontology mapping for HESRT metadata. Maps source-preserving normalized metadata to: - UBERON / Plant Ontology for anatomy/tissue - MONDO for disease - NCBI Taxonomy for organism Only unique lexical identity matches and a small set of source-validated aliases are accepted automatically. Original metadata values are never overwritten. """ from __future__ import annotations import argparse import hashlib import json import re import unicodedata from collections import Counter from pathlib import Path from typing import Any import pandas as pd ROOT = Path(__file__).resolve().parents[2] DEFAULT_RELEASE = ROOT / "standardized_data/releases/hesrt-hf-20260917-v1" DEFAULT_WORK = DEFAULT_RELEASE / "work/ontology_mapping_v1" DEFAULT_CATALOG = DEFAULT_RELEASE / "public/releases/2026-09-17/catalog.parquet" CONTROL_OR_NON_DISEASE = { "normal": "control label", "healthy": "control label", "control": "control label", "reference": "control/reference label", "non-diabetic": "control/risk-state label", "non-diabetic single autoantibody positive": "risk-state label, not a disease identity", "non-diabetic multiple autoantibody positive": "risk-state label, not a disease identity", "ngt": "normal glucose tolerance phenotype", "cci": "controlled cortical impact procedure/injury model", "ir surgery": "intervention label", } CURATED_DISEASE: dict[str, dict[str, str]] = { "MS": { "id": "MONDO:0005301", "label": "multiple sclerosis", "reason": "GSE174647 title/summary explicitly expand MS as multiple sclerosis", "source": "NCBI GEO GSE174647 checked 2026-09-18", }, "CD": { "id": "MONDO:0005011", "label": "Crohn disease", "reason": "GSE283945 title/summary explicitly describe Crohn's disease", "source": "NCBI GEO GSE283945 checked 2026-09-18", }, "cSCC": { "id": "MONDO:0002529", "label": "skin squamous cell carcinoma", "reason": "GSE144239 title/summary explicitly describe cutaneous squamous cell carcinoma", "source": "NCBI GEO GSE144239 checked 2026-09-18", }, } CURATED_TISSUE: dict[str, dict[str, str]] = { "brain cortex": { "id": "UBERON:0000956", "label": "cerebral cortex", "reason": "UBERON lists 'brain cortex' as a related synonym of cerebral cortex", }, "entorhinal cortex (ERC)": { "id": "UBERON:0002728", "label": "entorhinal cortex", "reason": "parenthetical acronym retained in source label; anatomy term is explicit", }, "dorsal root ganglia": { "id": "UBERON:0000044", "label": "dorsal root ganglion", "reason": "plural source label corresponds to UBERON dorsal root ganglion", }, "mouse brain": { "id": "UBERON:0000955", "label": "brain", "reason": "species modifier stripped; sample organism is checked separately", }, "human coronary artery": { "id": "UBERON:0001621", "label": "coronary artery", "reason": "species modifier stripped; anatomy term is explicit", }, "normal colon": { "id": "UBERON:0001155", "label": "colon", "reason": "control-state modifier stripped; anatomy term is explicit", }, "Skin punch biopsy": { "id": "UBERON:0000014", "label": "zone of skin", "reason": "biopsy specimen explicitly identifies skin as anatomical source", }, "coronal brain section": { "id": "UBERON:0000955", "label": "brain", "reason": "section/orientation modifiers stripped; anatomy term is explicit", }, "Coronal brain section": { "id": "UBERON:0000955", "label": "brain", "reason": "section/orientation modifiers stripped; anatomy term is explicit", }, "left hemisphere mice brain": { "id": "UBERON:0000955", "label": "brain", "reason": "laterality/species modifiers stripped; anatomy term is explicit", }, "Lumbar spinal cord (L3/L4)": { "id": "UBERON:0002240", "label": "spinal cord", "reason": "regional modifier stripped; spinal cord is explicit", }, "Postmortem temporal cortex": { "id": "UBERON:0016538", "label": "temporal cortex", "reason": "postmortem modifier stripped; anatomy term is explicit", }, "Whole skin": { "id": "UBERON:0000014", "label": "zone of skin", "reason": "whole-specimen modifier stripped; skin is explicit", }, "brain (hypothalamus)": { "id": "UBERON:0001898", "label": "hypothalamus", "reason": "more specific anatomy is explicit in parentheses", }, } # Frozen exact NCBI Taxonomy lookups, queried with the Scientific Name field on # 2026-09-18. Multi-species values are represented as multiple IDs. NCBITAXON = { "Ambystoma mexicanum": ["8296"], "Anolis carolinensis": ["28377"], "Canis lupus familiaris": ["9615"], "Ciona intestinalis": ["7719"], "Columba livia": ["8932"], "Dugesia japonica": ["6161"], "Electrophorus voltai": ["2609070"], "Gallus gallus": ["9031"], "Homo sapiens": ["9606"], "Homo sapiens; Mus musculus": ["9606", "10090"], "Macaca mulatta": ["9544"], "Mchenga conophoros": ["35575"], "Medicago truncatula": ["3880"], "Mesocricetus auratus": ["10036"], "Mus musculus": ["10090"], "Oryctolagus cuniculus": ["9986"], "Rattus norvegicus": ["10116"], "Sus scrofa": ["9823"], "Taeniopygia guttata": ["59729"], "Takifugu alboplumbeus": ["433688"], "Xenopus laevis": ["8355"], "human skin metagenome": ["539655"], } def sha256_file(path: Path) -> str: h = hashlib.sha256() with path.open("rb") as handle: for block in iter(lambda: handle.read(8 * 1024**2), b""): h.update(block) return h.hexdigest() def norm(value: str) -> str: text = unicodedata.normalize("NFKC", str(value)).casefold() text = text.replace("’", "'").replace("–", "-").replace("—", "-") return " ".join(text.split()) def lexical_norm(value: str) -> str: """Normalize punctuation without erasing biologically meaningful +/- signs.""" text = unicodedata.normalize("NFKD", str(value)).casefold().replace("’", "'") text = re.sub(r"(?<=\w)-(?=\w)", " ", text) text = text.replace("'", "") text = re.sub(r"[^\w+\-]+", " ", text) return " ".join(text.split()) def parse_obo(path: Path, prefix: str, ontology: str) -> dict[str, Any]: terms: dict[str, dict[str, Any]] = {} current: dict[str, Any] = {} def flush() -> None: if ( current.get("id", "").startswith(prefix) and current.get("name") and not current.get("obsolete") ): terms[current["id"]] = { "id": current["id"], "name": current["name"], "synonyms": list(current.get("synonyms", [])), "ontology": ontology, } with path.open(errors="replace") as handle: for raw in handle: line = raw.rstrip("\n") if line == "[Term]": flush() current = {} continue if line.startswith("["): flush() current = {} continue if line.startswith("id: "): current["id"] = line[4:] elif line.startswith("name: "): current["name"] = line[6:] elif line == "is_obsolete: true": current["obsolete"] = True elif line.startswith("synonym: "): match = re.match(r'synonym: "(.*?)" (EXACT|BROAD|NARROW|RELATED)', line) if match: current.setdefault("synonyms", []).append( {"text": match.group(1), "scope": match.group(2)} ) flush() exact: dict[str, set[str]] = {} lexical: dict[str, set[str]] = {} for term_id, term in terms.items(): entries = [(term["name"], "label")] + [ (item["text"], "exact_synonym") for item in term["synonyms"] if item["scope"] == "EXACT" ] for text, _ in entries: exact.setdefault(norm(text), set()).add(term_id) lexical.setdefault(lexical_norm(text), set()).add(term_id) return { "terms": terms, "exact": exact, "lexical": lexical, "path": str(path), "sha256": sha256_file(path), } def choose_unique(index: dict[str, set[str]], key: str) -> str | None: ids = index.get(key, set()) return next(iter(ids)) if len(ids) == 1 else None def ontology_match( value: Any, ontology: dict[str, Any], *, allow_lexical: bool = False, ) -> dict[str, Any]: if value is None or (isinstance(value, float) and pd.isna(value)): return { "id": None, "label": None, "ontology": None, "status": "missing_source_value", "match_type": None, "note": None, } text = str(value).strip() exact_ids = ontology["exact"].get(norm(text), set()) exact_id = next(iter(exact_ids)) if len(exact_ids) == 1 else None if exact_id: term = ontology["terms"][exact_id] match_type = "exact_label" if norm(term["name"]) == norm(text) else "exact_synonym" return { "id": exact_id, "label": term["name"], "ontology": term["ontology"], "status": "mapped", "match_type": match_type, "note": None, } if len(exact_ids) > 1: return { "id": None, "label": None, "ontology": None, "status": "ambiguous", "match_type": "ambiguous_exact", "note": "; ".join(sorted(exact_ids)), } if allow_lexical: lexical_ids = ontology["lexical"].get(lexical_norm(text), set()) lexical_id = next(iter(lexical_ids)) if len(lexical_ids) == 1 else None if lexical_id: term = ontology["terms"][lexical_id] return { "id": lexical_id, "label": term["name"], "ontology": term["ontology"], "status": "mapped", "match_type": "punctuation_normalized_exact", "note": None, } if len(lexical_ids) > 1: return { "id": None, "label": None, "ontology": None, "status": "ambiguous", "match_type": "ambiguous_punctuation_normalized_exact", "note": "; ".join(sorted(lexical_ids)), } return { "id": None, "label": None, "ontology": None, "status": "unmapped", "match_type": None, "note": None, } def map_tissue(value: Any, organism: Any, uberon: dict[str, Any], po: dict[str, Any]) -> dict[str, Any]: if value is None or (isinstance(value, float) and pd.isna(value)): return ontology_match(value, uberon) text = str(value).strip() org = "" if organism is None or pd.isna(organism) else str(organism) # Plant Ontology is preferred for known plant samples. if org == "Medicago truncatula": result = ontology_match(text, po, allow_lexical=True) if result["status"] == "mapped": return result result = ontology_match(text, uberon, allow_lexical=True) if result["status"] == "mapped": return result if text in CURATED_TISSUE: item = CURATED_TISSUE[text] # Guard species-specific rewrites against incompatible samples. if text == "mouse brain" and org != "Mus musculus": return result if text == "human coronary artery" and org != "Homo sapiens": return result return { "id": item["id"], "label": item["label"], "ontology": "UBERON", "status": "mapped", "match_type": "curated_anatomy_phrase", "note": item["reason"], } return result def map_disease(value: Any, organism: Any, mondo: dict[str, Any]) -> dict[str, Any]: if value is None or (isinstance(value, float) and pd.isna(value)): return ontology_match(value, mondo) text = str(value).strip() key = norm(text) if key in CONTROL_OR_NON_DISEASE: return { "id": None, "label": None, "ontology": None, "status": "control_or_non_disease", "match_type": "curated_non_disease_state", "note": CONTROL_OR_NON_DISEASE[key], } result = ontology_match(text, mondo, allow_lexical=True) if result["status"] == "mapped": return result if text in CURATED_DISEASE: item = CURATED_DISEASE[text] return { "id": item["id"], "label": item["label"], "ontology": "MONDO", "status": "mapped", "match_type": "source_validated_alias", "note": item["reason"] + "; " + item["source"], } # A few source labels carry an explicit parenthetical acronym. Only strip # a frozen allow-list where the remaining full phrase is an ontology term. parenthetical = { "Psoriasis (PsO)": "psoriasis", "Biliary tract cancer (BTC)": "biliary tract cancer", "Diabetic Kidney Disease (DKD)": "diabetic kidney disease", "Psoriatic Arthritis (PsA)": "psoriatic arthritis", "Acute Kidney Injury (AKI)": "acute kidney injury", } if text in parenthetical: base = parenthetical[text] base_result = ontology_match(base, mondo, allow_lexical=True) if base_result["status"] == "mapped": base_result["match_type"] = "parenthetical_acronym_stripped_exact" base_result["note"] = f"mapped full disease phrase after stripping source acronym from {text!r}" return base_result # Human context resolves the single observed MONDO ambiguity for multiple # myeloma: the alternative term is explicitly non-human-animal. if text.casefold() == "multiple myeloma" and str(organism) == "Homo sapiens": return { "id": "MONDO:0009693", "label": "plasma cell myeloma", "ontology": "MONDO", "status": "mapped", "match_type": "human_context_disambiguation", "note": "human sample; rejected MONDO non-human-animal alternative", } return result def map_organism(value: Any) -> dict[str, Any]: if value is None or (isinstance(value, float) and pd.isna(value)): return { "ids": None, "labels": None, "status": "missing_source_value", "match_type": None, } text = str(value).strip() ids = NCBITAXON.get(text) if not ids: return {"ids": None, "labels": None, "status": "unmapped", "match_type": None} labels = [part.strip() for part in text.split(";")] return { "ids": "; ".join(f"NCBITaxon:{taxid}" for taxid in ids), "labels": "; ".join(labels), "status": "mapped", "match_type": "exact_scientific_name" if len(ids) == 1 else "multi_taxon_exact", } def main() -> None: parser = argparse.ArgumentParser(description=__doc__) parser.add_argument("--catalog", type=Path, default=DEFAULT_CATALOG) parser.add_argument("--work-dir", type=Path, default=DEFAULT_WORK) parser.add_argument("--output-catalog", type=Path) args = parser.parse_args() catalog_path = args.catalog.resolve() work = args.work_dir.resolve() ont = work / "ontologies" work.mkdir(parents=True, exist_ok=True) uberon = parse_obo(ont / "uberon.obo", "UBERON:", "UBERON") mondo = parse_obo(ont / "mondo.obo", "MONDO:", "MONDO") po = parse_obo(ont / "po.obo", "PO:", "PO") frame = pd.read_parquet(catalog_path) before_columns = list(frame.columns) before_gsm = frame["gsm_id"].astype(str).tolist() output_rows: list[dict[str, Any]] = [] for row in frame.itertuples(index=False): tissue = map_tissue(row.tissue, row.organism, uberon, po) disease = map_disease(row.disease, row.organism, mondo) organism = map_organism(row.organism) output_rows.append({ "gsm_id": row.gsm_id, "tissue_ontology_id": tissue["id"], "tissue_ontology_label": tissue["label"], "tissue_ontology": tissue["ontology"], "tissue_ontology_status": tissue["status"], "tissue_ontology_match_type": tissue["match_type"], "tissue_ontology_note": tissue["note"], "disease_ontology_id": disease["id"], "disease_ontology_label": disease["label"], "disease_ontology": disease["ontology"], "disease_ontology_status": disease["status"], "disease_ontology_match_type": disease["match_type"], "disease_ontology_note": disease["note"], "organism_ontology_ids": organism["ids"], "organism_ontology_labels": organism["labels"], "organism_ontology_status": organism["status"], "organism_ontology_match_type": organism["match_type"], }) mapping = pd.DataFrame(output_rows) assert len(mapping) == len(frame) == 4468 assert mapping.gsm_id.is_unique assert mapping.gsm_id.astype(str).tolist() == before_gsm merged = frame.merge(mapping, on="gsm_id", how="left", validate="one_to_one", sort=False) assert merged[before_columns].equals(frame[before_columns]) mapping.to_parquet(work / "ontology_mapping.parquet", index=False, compression="zstd") mapping.to_csv(work / "ontology_mapping.csv", index=False) merged.to_parquet(work / "catalog.ontology-mapped.parquet", index=False, compression="zstd") merged.to_csv(work / "catalog.ontology-mapped.csv", index=False) unresolved_rows: list[dict[str, Any]] = [] for field, status_col in [ ("tissue", "tissue_ontology_status"), ("disease", "disease_ontology_status"), ("organism", "organism_ontology_status"), ]: tmp = merged[[field, status_col]].copy() tmp = tmp[tmp[field].notna()] grouped = tmp.groupby([field, status_col], dropna=False).size().reset_index(name="samples") for item in grouped.to_dict("records"): if item[status_col] != "mapped": unresolved_rows.append({ "field": field, "value": item[field], "status": item[status_col], "samples": int(item["samples"]), }) unresolved = pd.DataFrame(unresolved_rows) if not unresolved.empty: unresolved = unresolved.sort_values(["field", "samples", "value"], ascending=[True, False, True]) unresolved.to_csv(work / "unresolved_terms.csv", index=False) summary = { "schema_version": "hesrt-ontology-mapping-v1", "input_catalog": str(catalog_path), "input_catalog_sha256": sha256_file(catalog_path), "samples": len(frame), "ontologies": { "UBERON": {"data_version": "releases/2026-06-19", "sha256": uberon["sha256"]}, "MONDO": {"data_version": "releases/2026-09-01", "sha256": mondo["sha256"]}, "PO": {"data_version": "releases/2026-01-09", "sha256": po["sha256"]}, "NCBITaxon": { "lookup": "NCBI E-utilities taxonomy Scientific Name exact queries", "queried": "2026-09-18", "frozen_names": len(NCBITAXON), }, }, "coverage": { "tissue": dict(Counter(merged.tissue_ontology_status)), "disease": dict(Counter(merged.disease_ontology_status)), "organism": dict(Counter(merged.organism_ontology_status)), }, "match_types": { "tissue": dict(Counter(merged.tissue_ontology_match_type.dropna())), "disease": dict(Counter(merged.disease_ontology_match_type.dropna())), "organism": dict(Counter(merged.organism_ontology_match_type.dropna())), }, "invariants": { "row_count_preserved": len(merged) == len(frame), "gsm_order_preserved": merged.gsm_id.astype(str).tolist() == before_gsm, "original_columns_bytewise_dataframe_equal": bool(merged[before_columns].equals(frame[before_columns])), "original_values_overwritten": False, }, } (work / "summary.json").write_text(json.dumps(summary, indent=2, ensure_ascii=False) + "\n") provenance = { "schema_version": "hesrt-ontology-provenance-v1", "mapping_policy": ( "Unique ontology label/EXACT synonym matches are accepted. Punctuation-only lexical " "normalization is accepted only when unique. A small frozen set of anatomy phrases and " "three disease abbreviations are curated with explicit reasons/source checks. Controls, " "procedures and non-disease states are not forced into MONDO. No fuzzy candidate is " "accepted automatically." ), "curated_tissue": CURATED_TISSUE, "curated_disease": CURATED_DISEASE, "control_or_non_disease": CONTROL_OR_NON_DISEASE, "ncbitaxon": NCBITAXON, "ontologies": summary["ontologies"], "script_sha256": sha256_file(Path(__file__)), } (work / "ontology_provenance.json").write_text( json.dumps(provenance, indent=2, ensure_ascii=False) + "\n" ) if args.output_catalog: out = args.output_catalog.resolve() out.parent.mkdir(parents=True, exist_ok=True) merged.to_parquet(out, index=False, compression="zstd") print(json.dumps(summary, indent=2, ensure_ascii=False)) if __name__ == "__main__": main()