config.yaml

# Information about Babel
babel:
  github_url: "https://github.com/NCATSTranslator/Babel"

# Build information. Currently unstructured -- you can use this to write down any notes on what is going on with
# this specific build of Babel.
build:
  branch: babel-1.17
  workarounds:
  - DrugBank downloads are currently disabled https://github.com/cthoyt/drugbank-downloader/issues/23 -- I used the previously downloaded 5-1-13.

# Versions that need to be updated on every release.
biolink_version: "b0d9ef6494af9b3ab931e9505d446ca8c212f50f" # The next release after Biolink 4.4.2
umls_version: "2026AA"
rxnorm_version: "06012026"
drugbank_version: "5-1-13" # Really 5-1-21, but since downloads are currently not allowed, we need to reuse our most recent download.

# Overall inputs and outputs.
input_directory: input_data
download_directory: babel_downloads
intermediate_directory: babel_outputs/intermediate
output_directory: babel_outputs
# Where DuckDB spills larger-than-memory intermediates. setup_duckdb() gives each job its own
# duckdb-$SLURM_JOB_ID subdirectory here. Override per run with the BABEL_DUCKDB_TEMP_DIR
# environment variable; see slurm/README.md ("Temporary Scratch Space") for the trade-offs.
tmp_directory: babel_downloads/tmp

# HTTP settings.
http: {}

# SPARQL query settings.
sparql:
  # Total number of attempts (initial try + retries) before giving up; must be >= 1.
  # Example: max_attempts=3 → one initial try, then up to two retries.
  max_attempts: 3
  # Base delay in seconds between attempts; must be >= 0. Actual delay grows exponentially
  # (attempt 1 → 1 s, attempt 2 → 2 s, attempt 3 → 4 s, …). Accepts fractional values.
  retry_base_delay_seconds: 1

# Maps Python compendium names (as used in src/createcompendia/ and tests) to the
# directory names that Snakemake uses under intermediate_directory.
# Compendia not listed here use their own name as the directory name.
# Update this whenever a new semantic type uses a shortened or different directory name.
compendium_directories:
  diseasephenotype: disease
  processactivitypathway: process

#
# SHARED
#

# DuckDB settings for use in all DuckDB connections.
# write_buffer_row_group_count=1: flush Parquet row groups to disk eagerly. The default of 5
# means DuckDB holds 5 row groups × threads × ~4 GiB in write buffers before flushing (~76 GiB
# peak at 128 GB allocation). Setting it to 1 keeps write-buffer RAM proportional to one row
# group, at no cost to compression ratio or query performance.
#
# enable_external_file_cache=false: DuckDB's external file cache keeps Parquet file blocks mmapped
# across queries, which is one source of memory mappings. The cross-compendium report rules were
# hitting the kernel's per-process mapping limit (vm.max_map_count, default 65530) and failing with
# `bad allocation` on a small allocation despite free RAM -- an address-space (mmap-count) limit.
# Disabling this cache removes it as a mapping source, but note it was NOT sufficient on its own:
# the dominant source is DuckDB's buffer pool (its allocator retains ~1.3 MB mappings up to the
# query's peak memory), so the report rules also cap memory_limit far below their natural peak. The
# definitive fix is for the cluster to raise vm.max_map_count (issue #846); until then we keep this
# off (it only costs a re-read from disk for our read-once/scan-twice queries, no correctness change).
duckdb_config:
  write_buffer_row_group_count: 1
  enable_external_file_cache: false

#
# UMLS
#
umls:
  subset: "full"
  # Replace with "level-0" to get only the level 0 subset (https://www.nlm.nih.gov/research/umls/licensedcontent/umlsknowledgesources.html).
  # subset: "level-0"

#
# PROTEINS
#

# Chris Bizon prepared a list of UMLS/UniProtKB mappings which we download and use.
UMLS_UniProtKB_download_raw_url: "https://raw.githubusercontent.com/cbizon/UMLS_UniProtKB/refs/heads/main/outputs/UMLS_UniProtKB.tsv"

#
# The rest of these configs need to be cleaned up.
#

ncbi_files:
  - gene2ensembl.gz
  - gene_info.gz
  - gene_orthologs.gz
  - gene_refseq_uniprotkb_collab.gz
  - mim2gene_medgen

generate_dirs_for_labels_and_synonyms_prefixes: [GO, CL, NCIT, UBERON, CHEBI, HP, MONDO, PR]

# TODO: This doesn't appear to be used anywhere -- but it might be neat to use this list to make sure we got all of them.
ubergraph_ontologies: [UBERON, CL, GO, NCIT, ECO, ECTO, ENVO, HP, UPHENO, BFO, BSPO, CARO, CHEBI, CP, GOREL, IAO, MAXO, MONDO, PATO, PR, RO, UBPROP]

mods: [WormBase, FB, MGI, ZFIN, RGD, SGD]

common:
  labels: [ubergraph/labels]
  synonyms: [ubergraph/synonyms.jsonl]
  descriptions: [ubergraph/descriptions.jsonl]

anatomy_prefixes:  [UBERON, GO, CL, UMLS, MESH, NCIT, SNOMEDCT]
anatomy_ids:       [UBERON, GO, CL, UMLS, MESH, NCIT]
anatomy_concords:  [UBERON, GO, CL, UMLS, WIKIDATA]
anatomy_outputs:   [AnatomicalEntity.txt, Cell.txt, CellularComponent.txt, GrossAnatomicalStructure.txt]
# Prefixes whose identifiers are treated as globally unique within a clique: if merging two
# cliques would produce a clique containing more than one identifier with one of these prefixes,
# the merge is blocked.  UBERON and GO are authoritative enough that two distinct terms from
# either ontology should never be collapsed into the same equivalence set.  If you add a new
# anatomy source whose concord file links to UBERON or GO, verify that it doesn't introduce
# spurious merges before removing either prefix from this list.
anatomy_unique_prefixes: [UBERON, GO]

gene_labels:   [HGNC, NCBIGene, UMLS, OMIM]
gene_ids:      [ENSEMBL, HGNC, NCBIGene, UMLS, OMIM, ZFIN, WormBase, FB, MGI, RGD, SGD]
gene_concords: [NCBIGene, NCBIGeneENSEMBL, medgen, UMLS, UMLS_NCBIGene]
gene_outputs:  [Gene.txt]

protein_labels:   [UniProtKB, PR, UMLS]
protein_synonyms: [PR, UMLS]
protein_ids:      [ENSEMBL, MESH, UniProtKB, PR, UMLS]
protein_concords: [UniProtKB, PR, NCIT_UniProtKB, NCIT_UMLS, UMLS, UMLS_UniProtKB]
protein_outputs:  [Protein.txt]

disease_labelsandsynonyms: [MONDO, DOID, Orphanet, HP, MESH, NCIT, UMLS, SNOMEDCT, EFO]
disease_ids:               [MONDO, DOID, Orphanet, HP, MESH, NCIT, UMLS, OMIM, EFO]
disease_concords:          [HP, MONDO, UMLS, DOID, EFO, Manual]
disease_outputs:           [Disease.txt, PhenotypicFeature.txt]

process_labels:   [GO, REACT, RHEA, EC, SMPDB, PANTHER.PATHWAY, UMLS]
process_ids:      [GO, REACT, RHEA, EC, SMPDB, PANTHER.PATHWAY, UMLS]
process_concords: [GO, RHEA, UMLS]
process_outputs:  [Pathway.txt, BiologicalProcess.txt, MolecularActivity.txt]

unichem_datasources: [CHEMBL.COMPOUND, DRUGBANK, GTOPDB, CHEBI, UNII, HMDB, PUBCHEM.COMPOUND, DrugCentral]  # KEGG.COMPOUND removed from UniChem — https://github.com/NCATSTranslator/Babel/issues/834

chemical_labels:    [CHEMBL.COMPOUND, GTOPDB, CHEBI, UNII, HMDB, PUBCHEM.COMPOUND, DrugCentral, UMLS, DRUGBANK]  # KEGG.COMPOUND removed from UniChem — https://github.com/NCATSTranslator/Babel/issues/834
chemical_synonyms:  [GTOPDB, CHEBI, UNII, HMDB, PUBCHEM.COMPOUND, UMLS, DRUGBANK]
chemical_concords:  [wikipedia_mesh_chebi, PUBCHEM_MESH, mesh_cas, mesh_unii, PUBCHEM_CAS, GTOPDB, CHEBI, UMLS, DrugCentral, RXNORM]
chemical_ids:       [CHEMBL.COMPOUND, GTOPDB, CHEBI, UNII, HMDB, PUBCHEM.COMPOUND, DrugCentral, DRUGBANK, MESH, UMLS, RXNORM]  # KEGG.COMPOUND removed from UniChem — https://github.com/NCATSTranslator/Babel/issues/834
chemical_outputs:   [MolecularMixture.txt, SmallMolecule.txt, Polypeptide.txt, ComplexMolecularMixture.txt, ChemicalEntity.txt, ChemicalMixture.txt, Drug.txt]

drugchemicalconflated_synonym_outputs: [DrugChemicalConflated.txt]
geneproteinconflated_synonym_outputs:  [GeneProteinConflated.txt]

taxon_labels:   [NCBITaxon, MESH, UMLS]
taxon_synonyms: [NCBITaxon, UMLS]
taxon_ids:      [NCBITaxon, MESH, UMLS]
taxon_concords: [NCBI_MESH, UMLS]
taxon_outputs:  [OrganismTaxon.txt]

cell_line_outputs: [CellLine.txt]

genefamily_labels:  [PANTHER.FAMILY, HGNC.FAMILY]
genefamily_ids:     [PANTHER.FAMILY, HGNC.FAMILY]
genefamily_outputs: [GeneFamily.txt]

umls_outputs: [umls.txt]

macromolecularcomplex_outputs: [MacromolecularComplex.txt]

ubergraph_iri_stem_to_prefix_map:
  "https://identifiers.org/ncbigene/": NCBIGene
  "http://www.ncbi.nlm.nih.gov/gene/": NCBIGene
  "http://www.genenames.org/cgi-bin/gene_symbol_report?hgnc_id=": HGNC
  "http://www.informatics.jax.org/marker/MGI:": MGI
  "http://www.pombase.org/spombe/result/": PomBase
  "http://www.wormbase.org/species/c_elegans/gene/": WormBase
  "http://flybase.org/reports/": FB
  "http://www.ecogene.org/gene/": ecogene
  "http://zfin.org/action/marker/view/": ZFIN
  "http://birdgenenames.org/cgnc/GeneReport?id=": cgnc
  "http://www.ensemblgenomes.org/id/": ENSEMBL
  "http://www.ensembl.org/id/": ENSEMBL
  "http://purl.obolibrary.org/obo/AISM_": AISM
  "http://purl.obolibrary.org/obo/BCO_": BCO
  "http://purl.obolibrary.org/obo/CDNO_": CDNO
  "http://purl.obolibrary.org/obo/CLAO_": CLAO
  "http://purl.obolibrary.org/obo/GNO_": GNO
  "http://purl.obolibrary.org/obo/HAO_": HAO
  "http://purl.obolibrary.org/obo/LEPAO_": LEPAO
  "http://purl.obolibrary.org/obo/MMO_": MMO
  "http://purl.obolibrary.org/obo/MRO_": MRO
  "http://purl.obolibrary.org/obo/MmusDv_": MmusDv
  "http://purl.obolibrary.org/obo/OARCS_": OARCS
  "http://purl.obolibrary.org/obo/OBA_": OBA
  "http://purl.obolibrary.org/obo/PCL_": PCL
  "http://purl.obolibrary.org/obo/PECO_": PECO
  "http://purl.obolibrary.org/obo/PPO_": PPO
  "http://purl.obolibrary.org/obo/TS_": TS
  "http://purl.obolibrary.org/obo/UO_": UO
  "http://purl.obolibrary.org/obo/PCO_": PCO
  "http://rgd.mcw.edu/rgdweb/report/gene/main.html?id=": RGD

publication_outputs: [Publication.txt]

geneprotein_outputs: [GeneProtein.txt]

drugchemical_outputs: [DrugChemical.txt]

preferred_name_boost_prefixes:
  biolink:ChemicalEntity: [DRUGBANK, DrugCentral, CHEBI, MESH, GTOPDB]

ensembl_datasets_to_skip:
#  - elucius_gene_ensembl
#  - hgfemale_gene_ensembl
#  - charengus_gene_ensembl
#  - otshawytscha_gene_ensembl
#  - aocellaris_gene_ensembl
  - omykiss_gene_ensembl        # 2026jun5: keeps getting stuck with this dataset for some reason.

# Labels longer than this limit are demoted (not used as preferred label if a shorter alternative exists).
# Keyed by Biolink type; types not listed here are never demoted. Uses ancestor traversal, so
# biolink:ChemicalEntity applies to all chemical subtypes (SmallMolecule, Drug, etc.).
# See https://github.com/NCATSTranslator/Babel/issues/597
demote_labels_longer_than:
  biolink:ChemicalEntity: 25