Three corpus sources explain empty or sparse Pfam results from bakta in different ways. Pfam is a database of protein-domain families, and bakta is a bacterial genome annotation pipeline that can report Pfam hits. The explanations are a query-format problem, a reduced profile set, and a search restricted to hypothetical proteins; one source also says the correct query format is still unestablished. The disagreement matters because a Pfam table with no hits is negative homology-search evidence. As homology-search-negative-evidence argues, such a result does not by itself demonstrate that a domain is absent. The 12/22 coverage gap 1 and the 0 hits across all 11 domains tested 2 are the concrete cases. No source tests another’s mechanism, so the explanations stand unreconciled. 1,3,2
Evidence Sides
Side A: Query format, then a reduced “core” profile set (plant microbiome report) The plant microbiome report treats its original zero as a resolved query-format problem involving versioned IDs (Pfam accessions that carry a version suffix). 1 It attributes the separate 12/22 coverage gap to a likely reduced “core” Pfam profile set in bakta. 1 Because the report calls the profile-set explanation “likely”, it is an inference rather than a demonstrated cause. 1
Side B: Pfam search run only on hypotheticals (discoveries digest) The discoveries digest attributes the under-reporting to bakta running its Pfam search only on hypotheticals, meaning proteins without another assigned function. 3
Side C: Correct query format not yet established (pitfalls digest) The pitfalls digest’s entry reports 0 hits across all 11 domains tested and still describes the correct query format as unestablished. 2 The plant microbiome report, by contrast, treats its versioned-ID format problem as resolved. 1
Possible Reconciliations
- Hypothesis 1: The mechanisms may stack. A versioned-ID format error could explain total zeros, while a hypothetical-only search or a reduced profile set could explain partial coverage gaps such as the 12/22 gap. No source has tested this. 1,3,2
- Hypothesis 2: The pitfalls entry may predate or not reflect the plant microbiome report’s format fix, so Side C would describe an earlier state of knowledge rather than a live contradiction. The passages do not establish their relative timing, so this remains untested.
- Hypothesis 3: A reduced “core” profile set and a hypothetical-only search are not mutually exclusive; both could reduce coverage at once.
Resolving Work
- Re-run the pitfalls test across the 11 domains, querying the bakta Pfam tables with both versioned and unversioned accessions, to show whether the 0-hit result is a format artifact and whether the plant microbiome fix generalizes.
- For the items behind the 12/22 coverage gap, compare bakta Pfam output against an independent full-Pfam search on the same proteins, for example HMMER
hmmscan(which scans protein sequences against hidden Markov model profiles) with the complete Pfam library. Families found by the full library but absent from bakta’s profile set would test the reduced-profile-set explanation. - Split bakta Pfam hits by whether the protein is annotated as hypothetical. If no non-hypothetical protein ever carries a Pfam hit, that would support the discoveries digest’s mechanism.
- Inspect the bakta version and database configuration used in each project to show whether a “core” profile subset was installed and whether the three sources describe different software setups.