菜单
Identify and populate the kg_microbe_match field for CultureMech media records by finding exact ingredient matches in the KG-Microbe mediadive graph
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use this skill to quality-check CultureMech growth-media and solution recipes — verify schema/structure, MediaIngredientMech ingredient linkages and CHEBI mappings, solution references, data quality (placeholders, units, concentrations), and categorization. Use after creating/editing a recipe, before a KG export, or for periodic maintenance. Issues are graded P1 (blocking) → P4 (optional); P3/P4 can be auto-fixed.
Use this skill to manage CultureMech record identifiers — find the highest existing CultureMech id, mint the next id, and insert new media/solution records with correct id placement. Use when adding or importing recipe records, or reconciling id collisions.
Use this skill to integrate the MediaIngredientMech ingredient variant hierarchy (parent/child relationships — purity levels, hydrates, stereoisomers) into CultureMech recipe ingredient references. Use when linking recipe ingredients to their canonical MIM records or resolving variant forms.
Validate that every ontology ID carries its correct ontology LABEL across CultureMech recipe YAMLs and SSSOM mapping products. Uses LinkML schema bindings + linkml-term-validator for term.{id,label} (canonical label) and a shared OAK validator for products. Run report-then-enforce; triage drift as wrong-label vs wrong-id.
Scrape growth-medium recipes directly from the RIKEN JCM GRMD database and emit CultureMech-normalized YAML records. Use to close the gap between CultureMech's JCM coverage (originally ingested second-hand via MediaDive/TOGO) and the live JCM catalogue — e.g. when newly added high-GRMD media are missing, or to audit JCM completeness.
Generate an interactive UMAP scatter plot of CultureMech CHEBI ingredients in KG-Microbe embedding space — each point is one unique ingredient, sized by frequency and colored by occurrence tier
| name | match-kg-microbe |
| description | Identify and populate the kg_microbe_match field for CultureMech media records by finding exact ingredient matches in the KG-Microbe mediadive graph |
| category | enrichment |
| requires_database | false |
| requires_internet | false |
| version | 1.0.0 |
| tags | ["kg-microbe","matching","enrichment","ontology","mediadive"] |
Purpose: For each CultureMech media record, find the corresponding mediadive.medium:XXX node in the KG-Microbe knowledge graph by comparing CHEBI ingredient sets (concentrations ignored), and populate the kg_microbe_match field.
Why: Enables direct traceability between CultureMech recipes and the KG-Microbe graph, supporting cross-database queries, validation against authoritative DSMZ formulations, and knowledge graph integration.
Schema field: kg_microbe_match (pattern: ^mediadive\.medium:[0-9a-zA-Z_-]+$)
Match logic: Exact CHEBI ingredient set equality via the medium → solution → CHEBI graph path in edges.tsv.
term.id: CHEBI:XXXXX) to media ingredientsScans all 15,000+ CultureMech YAML files at once. Builds an in-memory CHEBI-set → mediadive.medium index from edges.tsv, then matches every CultureMech record.
# From the culturebotai-claw repo (dry run first)
python ../culturebotai-claw/scripts/match_culturemech_to_kg.py \
--culturemech . \
--kg-microbe ~/Documents/VIMSS/ontology/KG-Hub/KG-Microbe/kg-microbe \
--dry-run
# Apply
python ../culturebotai-claw/scripts/match_culturemech_to_kg.py \
--culturemech . \
--kg-microbe ~/Documents/VIMSS/ontology/KG-Hub/KG-Microbe/kg-microbe
Or via the justfile in culturebotai-claw:
cd ~/Documents/VIMSS/ontology/KG-Hub/KG-Microbe/culturebotai-claw
just match-culturemech-dry # preview
just match-culturemech # apply
Uses src/culturemech/match/kg_media_matcher.py and the enrich_with_kg_microbe_matches.py script. Processes one category directory at a time.
# From within CultureMech repo (activate venv first)
source .venv/bin/activate
# Dry run on one category
python scripts/enrich_with_kg_microbe_matches.py \
--kg-microbe-dir ~/Documents/VIMSS/ontology/KG-Hub/KG-Microbe/kg-microbe \
--recipe-dir data/normalized_yaml/bacterial \
--dry-run
# Apply to one category
python scripts/enrich_with_kg_microbe_matches.py \
--kg-microbe-dir ~/Documents/VIMSS/ontology/KG-Hub/KG-Microbe/kg-microbe \
--recipe-dir data/normalized_yaml/bacterial
# All categories (loop)
for cat in bacterial algae archaea fungal specialized; do
python scripts/enrich_with_kg_microbe_matches.py \
--kg-microbe-dir ~/Documents/VIMSS/ontology/KG-Hub/KG-Microbe/kg-microbe \
--recipe-dir data/normalized_yaml/$cat
done
To check whether one specific recipe has a KG match:
source .venv/bin/activate
python - <<'EOF'
from pathlib import Path
from culturemech.match import KGMediaMatcher
import yaml
kg_dir = Path.home() / "Documents/VIMSS/ontology/KG-Hub/KG-Microbe/kg-microbe"
matcher = KGMediaMatcher(kg_dir)
recipe = yaml.safe_load(Path("data/normalized_yaml/bacterial/lb_agar.yaml").read_text())
match = matcher.find_exact_match(recipe)
print(f"Match: {match}")
EOF
~/Documents/VIMSS/ontology/KG-Hub/KG-Microbe/kg-microbe/data/transformed_last9/mediadive/
edges.tsv — medium→solution→CHEBI relationshipsnodes.tsv — medium node names and IDsterm.id: CHEBI:XXXXX populated for matching to workAs of 2026-04-09 (last full run):
To increase coverage: run FEBA ontology enrichment (just feba-enrich-ontology in culturebotai-claw) to add more CHEBI IDs, then re-run this skill.
1. Check current coverage
grep -rl "kg_microbe_match" data/normalized_yaml/ | wc -l
2. Run dry-run to see what would be matched
python ../culturebotai-claw/scripts/match_culturemech_to_kg.py \
--culturemech . --kg-microbe ~/Documents/.../kg-microbe --dry-run
3. Apply matches
python ../culturebotai-claw/scripts/match_culturemech_to_kg.py \
--culturemech . --kg-microbe ~/Documents/.../kg-microbe
4. Verify a sample
grep "kg_microbe_match" data/normalized_yaml/bacterial/lb_agar.yaml
5. Commit
git add data/normalized_yaml/
git commit -m "Add kg_microbe_match for N media matched to KG-Microbe nodes"
A matched media file gets one new field, e.g.:
id: CultureMech:009373
name: lb_agar
kg_microbe_match: mediadive.medium:74
...
The value is always a mediadive.medium:XXX node ID. When multiple KG-Microbe media have identical CHEBI ingredient sets, the one with the lowest numeric ID is used.
No matches found: Check that term.id fields are populated in the media's ingredients. Unmapped ingredients cannot contribute to matching.
KG-Microbe path not found: Verify the --kg-microbe path points to the repo root containing data/transformed_last9/mediadive/edges.tsv.
Match seems wrong: Use Option C to inspect the specific recipe. The match is purely by CHEBI set equality — verify both the CultureMech and KG-Microbe recipes actually represent the same medium.
Many media unmatched: The KG-Microbe mediadive graph contains ~3,317 media (DSMZ-derived). CultureMech has 15,827. Most CultureMech media are not in the KG; low match rate is expected.
../culturebotai-claw/scripts/match_culturemech_to_kg.pyscripts/enrich_with_kg_microbe_matches.pysrc/culturemech/match/kg_media_matcher.pysrc/culturemech/schema/culturemech.yaml → kg_microbe_matchmatch-kg-microbe in MediaIngredientMech (uses kg_microbe_node_id field)docs/archive/KG_MICROBE_MATCHER_IMPLEMENTATION.mdjust feba-enrich-ontology in culturebotai-claw (adds CHEBI IDs to expand matchable media)