hmm_library

HMM Library

Efesto ships with a curated library of profile HMMs in hmm_library/. All models are tracked in a versioned registry (hmm_library/hmm_registry.tsv) with provenance, bitscore cutoffs, and curation status.

---

Sources

Source	Models (active)	Description
FeGenie	196	Original iron-cycling profiles from Garber et al. 2020
Tabuteau et al. 2025	130	Iron acquisition profiles from KOfam, FeGenie, and NCBI NF*
MetHMMDB	105	Metal mobility resistance gene HMMs (broad metal scope)
NCBIfam / TIGRFAM	21	Curated iron oxidation, regulation, sulfur assembly (SufA/B/C/D/S), molybdenum resistance, manganese, and Tad-pilus (CpaB) models
MnOxGeneTool	12	Manganese oxidation genes (Wang et al. 2025)
InterPro	8	Cross-referenced domain models
Curated (custom-built)	4	MtrA, MtoA subfamily rebuild, plus CbcL and DA_402 (MISO extracellular MHC) — built from literature-confirmed seed sequences, background-tested against the rest of the library
Pfam	2	Flp/Fap pilin (Tad pilus) and SufE, sourced directly from Pfam-A
Total active	478	After deduplication (176 deprecated; 654 total)

---

Categories

Iron cycling

Category	Description	Notable models
iron_oxidation	Enzymatic oxidation of Fe²⁺ → Fe³⁺	MtoA, MtrA, FoxABC, Cyc2, rusticyanin, cytochrome579
iron_reduction	Enzymatic reduction of Fe³⁺ → Fe²⁺	MtrA, MtrC, OmcS, OmcZ, DFE operons, CbcL, DA_402, Flp/CpaB (Tad pilus)
probable_iron_reduction	Fe reduction genes with lower specificity	CymA, omcE
possible_iron_oxidation_and_possible_iron_reduction	Dual-assignment before Mtr/Mto disambiguation	See MtrMto operon rule
iron_storage	Cellular iron storage proteins	Ferritin, bacterioferritin, Dps
iron_gene_regulation	Fe-responsive transcriptional regulators	Fur, DtxR, IscR, NsrR, PerR, SoxR
iron_stress	Iron-starvation biomarkers	Flavodoxin long, flavodoxin short
iron_sulfur_assembly	Fe-S cluster biosynthesis machinery	SufA, SufB, SufC, SufD, SufS, SufE, IscS
magnetosome_formation	Magnetosome island genes	MamA/B/E/K/P/M/Q/I/L/O

Iron acquisition (siderophore-centred)

Category	Description
iron_acquisition-siderophore_synthesis	NRPS and related enzymes; NRP/PK siderophore biosynthesis
iron_acquisition-siderophore_transport	Outer membrane TonB-dependent receptors for Fe-siderophore complexes
iron_acquisition-siderophore_transport_potential	TonB-ExbBD motor proteins; ABC transporter permease/ATPase subunits
iron_acquisition-heme_oxygenase	Enzymes releasing iron from heme
iron_acquisition-heme_transport	Outer membrane heme receptors, ABC transporters
iron_acquisition-iron_transport	ABC iron transporters (FbpABC, FeoABC, EfeU, FutABC, etc.)

Metal resistance (MetHMMDB)

Category	Description
metal_resistance-arsenic	Arsenate reductase, arsenite efflux (ArsABC, ArsH)
metal_resistance-chromium	Chromate efflux (ChrA)
metal_resistance-cobalt_zinc_cadmium	CzcABC RND efflux, ZnuABC uptake, ZinT, FieF, Zur
metal_resistance-copper	CopABCD, CusSR, CopY
metal_resistance-mercury	MerABCDEPRT
metal_resistance-molybdenum	ModABC ABC transporter, ModC ATPase (TIGRFAM)
metal_resistance-multimetal	CuAg/CdCoZn RND systems, broad-spectrum
metal_resistance-nickel	NikABCDE, NicT, CznABC
metal_resistance-silver	CuAg_CusA/C (overlapping CopA)
metal_resistance-non-specific	Broad-spectrum and non-metal genes (after curation)
metal_resistance-tellurium	TerABCDEZ, TehAB

---

Registry format (hmm_library/hmm_registry.tsv)

Tab-separated, one row per model. Columns:

Column	Description
stem	HMM file stem; must exactly match filename without .hmm
name	Human-readable gene name (shown in outputs)
accession	Pfam/TIGRFAM accession or empty
category	Functional category (directory name)
hmm_file	Relative path from hmm_library/
nseq	Training sequences used to build the HMM
cutoff	Calibrated bitscore cutoff (0 = zero-cutoff fallback)
date_added	ISO date
status	active or deprecated_*
reference	DOI or citation

Registry rule: stem must exactly match the HMM file's NAME field and the filename without .hmm. Mismatches silently break cutoff lookup and gene-name mapping. Run python scripts/curate_hmm_library.py --verify hmm_library/ to detect drift.

---

Bitscore thresholds

hmm_library/HMM-bitcutoffs.txt — two-column TSV: stem<TAB>bitscore.

Models	Threshold type
337 calibrated models	Per-HMM gathering cutoff (GA) from TIGRFAM/Pfam or manual calibration
129 zero-cutoff models	No calibrated threshold; use --zero_cutoff_min_bitscore (default 30.0)

Zero-cutoff models: MetHMMDB (115) + 14 FeGenie siderophore models. Raise --zero_cutoff_min_bitscore to reduce false positives from these models.

---

Gene name mapping

hmm_library/FeGenie-map.txt — two-column TSV: stem<TAB>gene_name.

Gene names from this file appear in the gene column of all output files. When a model has no entry, the raw stem is used as the gene name.

---

Deduplication layers

See Library curation log for full decision log.

Layer	Method	Deprecated
A — Minimum nseq	Source-specific thresholds (fegenie ≤ 5, methmmdb ≤ 10)	60
A — Name dedup	Same stem, keep higher nseq/coverage	88
B — Sequence dedup	MMseqs2 70% id / 80% cov on hmmemit consensus	12
Category mismatch	Manual review of cross-category hits	7

---

Startup provenance output

At startup, Efesto prints a provenance table listing active sources, model counts, and any models with nseq < 10 (flagged as low_training_data):

Source          Models  Calibrated  Zero-cutoff  nseq<10
FeGenie            196         182           14        0
Tabuteau           130          98           32        0
NCBIfam             15          15            0        0
MetHMMDB           115           0          115       10
────────────────────────────────────────────────────────
Total              456         295          161       10

WARNING: 10 models have nseq < 10 (limited training data):
  [methmmdb] CdCoZn_efflux_czcD_1 (nseq=7), ...

Models with nseq < 10 are included in runs but appear as low_confidence in output. Identify them via the model_nseq column in results-long.tsv.

---

Normalising legacy HMM formats

All models in the shipped library are in HMMER3/f format. If you add models built with HMMER < 3.1, run once with --normalize_hmms to convert them in-place.