What this entity is
Canonical name: Pfam (PFam protein-family database). 1
Known aliases: Pfam; PFam domains. 1
Stable external identifier: Not reported in the source documents. 1,2
Pfam is a domain-level annotation resource used to associate genes with protein-family domains and biological-process modules. 1
The NMDC context audit identifies Pfam/InterPro as the authority for nmdc.ref_data, a Pfam re-host containing 27,481 terms in the nmdc tenant. This refines the existing description by distinguishing Pfam’s annotation role from its separately maintained KBase Data Lakehouse reference-data copy. 2
Annotation and module interpretation
Pfam provided the broadest annotation coverage among the annotation sources evaluated for independent component analysis (ICA), a statistical decomposition method, of bacterial RB-TnSeq fitness data. 1
Adding Pfam domains and lowering the enrichment-overlap threshold from 3 to 2 increased the module annotation rate from 8% to 80%, expanded annotated modules from 92 to 890, and unlocked 7.6x more function predictions. 1
Pfam-based annotation supported module-level interpretation across 1,116 stable modules identified from 32 organisms, but the source does not attribute a separate number of those modules specifically to Pfam. 1
Pfam annotations were used alongside KEGG, SEED, and TIGRFam enrichment to generate 6,691 function predictions for hypothetical proteins across 32 organisms. 1
Pfam-based annotations provided broader coverage than KEGG KOs for module-level enrichment, because KEGG KOs were too gene-specific for that purpose. 1
The source cautions that Pfam operates at the domain level and may overcount functional associations. 1
The truly-dark-gene census refines this coverage claim by showing that Pfam rarely supplies domain-level evidence for genes that remain hypothetical after modern reannotation: only 4.0% of 5,870 unique truly dark gene clusters had Pfam hits, comprising 362 hits for 235 clusters. TPR repeats dominated non-DUF hits, while 56 clusters had DUF-only domains. 3
This result supports treating Pfam as a high-value source for partial functional clues without equating a domain hit with a complete gene-function assignment. Among truly dark genes, 4.6% had KEGG KOs, 43.5% had partial eggNOG-mapper signal, and 55.4% of COG assignments were in category S, meaning function unknown. 3
SNIPE domain architecture and phage-defense detection
The SNIPE-defense analysis refines Pfam’s role as a domain-level resource by correcting the domain signatures used for a specific phage-defense family: PF13250 is the correct Pfam assignment for DUF4041, while PF13455 (Mug113) is the SNIPE nuclease family and is distinct from canonical GIY-YIG PF01541. 4
In the surveyed KBase Data Lakehouse annotations, zero gene clusters contained both DUF4041/PF13250 and canonical GIY-YIG PF01541. Of 4,572 DUF4041-containing gene clusters, 54 carried the description “Meiotically up-regulated gene 113,” consistent with full-length SNIPE proteins containing the Mug113 nuclease annotation. This contradicts using PF01541 as the SNIPE nuclease marker and supports searching PF13250, DUF4041, or T5orf172 together with PF13455. 4
DUF4041/PF13250 occurred in 4,572 gene clusters across 1,696 species and 33 bacterial and archaeal phyla; 13.3% of these clusters were core, 30.7% accessory, and 56.1% singleton, making the accessory-plus-singleton fraction 86.7%. These results support interpreting SNIPE-associated Pfam domains as markers of a predominantly mobile defense-family distribution rather than stable core inheritance. 4
The SNIPE study supports Pfam’s use for broad defense-system screening through a curated signature list of approximately 20 Pfam accessions queried against the 833-million-row interproscan_domains table. 5
The phage-defense analysis refines the provenance description: InterProScan uses version-free accessions such as PF01867, whereas bakta_pfam_domains uses versioned accessions such as PF01867.29; cross-database joins therefore require stripping Bakta versions. 5
InterProScan filtering for analysis = 'Pfam' and the curated signature list returned ~500K rows in under 90 seconds, while the 18.8-million-row bakta_pfam_domains table contained 0 Cas1 PF01867 hits compared with ~25K in InterProScan. This supports provenance-aware resource selection rather than treating Pfam-backed tables as interchangeable. 5
The defense study also refines interpretation of Pfam-based detection: EggNOG description matching gave 96% CRISPR-Cas prevalence, whereas the specific Cas1 Pfam marker PF01867 gave approximately 55% on the same pangenomes. The 96% value is therefore treated as an upper bound, with detection-method labels required for CRISPR comparisons. 5
These results support the existing caution that domain-level annotation can overcount functional associations. The broad DrmB SNF2 helicase Pfam anchor PF00176 made DISARM accessory-enrichment results unreliable because PF00176 is also widespread in non-DISARM housekeeping helicases. 5
The SNIPE report further cautions that eggNOG Pfam annotations may miss divergent homologues and that DUF4041 may occur in non-SNIPE proteins; only 54/4,572 DUF4041 clusters showed Mug113 co-annotation. 4
Provenance and related interpretation
The audit’s identification of nmdc.ref_data as an external Pfam re-host supports maintaining provenance labels when interpreting this resource, especially because the nmdc name does not by itself establish NMDC ownership. 2
Pfam therefore supports process-level interpretation of cofitness-network-architecture and complements sequence-based gene-level approaches such as ortholog transfer, which achieved 95.8% strict precision, 91.2% coverage, and 0.934 F1 in the held-out benchmark. 1
The truly-dark-gene analysis supports this complementary role: 96.2% of the 6,427 truly dark genes had at least one clue, but only 4.0% had Pfam hits; Tier 3 genes with partial Pfam, COG, or eggNOG function and Tier 4 genes with phenotype-only evidence together comprised 2,314 genes prioritized for narrowing experimental hypotheses. 3
Related pages
- Pan-bacterial Fitness Modules via Independent Component Analysis — source summary for the fitness-module analysis. 1
- NMDC Context Audit — audit of provenance, scale, currency, and naming for NMDC-labeled KBase Data Lakehouse resources. 2
- Pan-Bacterial Anti-Phage Defense Arsenal — pan-bacterial defense-system detection and Pfam-based marker analysis. 5
- SNIPE Defense System in the KBase Data Lakehouse Pangenome — SNIPE domain architecture, prevalence, mobility, and phage-defense analysis. 4
- Truly Dark Genes — What Remains Unknown After Modern Annotation? — residual unknown genes after Bakta reannotation and functional-database integration. 3
- cofitness-network-architecture — cofitness modules and their network structure. 1
- gene-essentiality — distinction between process-level module context and gene-level function prediction. 1
- pangenome-integration — cross-organism alignment and conserved module families. 1
- provenance-aware-resource-discovery — provenance-aware interpretation of re-hosted reference resources. 2
- kegg — gene-level annotation source compared with Pfam. 1
- seed — annotation source used with Pfam for function predictions. 1
- tigrfam — annotation source used with Pfam for function predictions. 1
- pf13250 duf4041 — corrected Pfam family assignment for the SNIPE-associated DUF4041 domain. 4
- pf13455 mug113 — Pfam family assignment for the SNIPE nuclease domain. 4