SPKW Viral Clades Subproject

Parent project: SPKW.md

Overview

This subproject extends the Swiss-Prot Keyword (SPKW) review from single-organism and term-focused analyses to viruses across multiple viral clades. The goal is to separate three different phenomena that are conflated in raw SPKW annotations:

1. Legitimate viral biology - terms such as virion component, viral DNA genome replication, viral envelope, host-cell entry, or latency-replication decision can be appropriate for the right viral protein.
2. Taxon-dependent semantic mismatch - phage proteins should not be annotated with eukaryotic innate immune response terms when the host is a bacterium.
3. Specificity loss - many SPKW-derived annotations are directionally correct but too broad, especially generic enzymatic functions, generic membrane/virion components, and broad host-defense suppression terms.

The bacteriophage T4 subproject, SPKW-BPT4.md, is the first focused viral case study. This page is the virus-wide equivalent: it summarizes all viral SPKW candidates, folds in additional viral gene reviews, and defines clade-specific follow-up batches.

Summary Statistics

Database: ~/repos/go-db/db/virus.ddb
Queried: 2026-05-21

Distribution by Aspect

Naive SPKW-unique annotations:

Closure-filtered SPKW-unique annotations:

Swiss-Prot vs TrEMBL

Only 13.1% of virus SPKW annotations are on Swiss-Prot entries. This differs from human and T4, where the reviewed Swiss-Prot fraction is very high. For viruses as a whole, errors may come from either the KW->GO mapping layer or automatic keyword assignment on TrEMBL entries.

ViralZone-Primary Keyword Stratification

ViralZone pages are integrated with UniProt keywords, but the safest machine-readable mapping is the page-header data-about="kw:KW-####" field, not the visible "Links / Uniprot Keyword" box. A ViralZone sitemap audit on 2026-05-21 found 147 pages with exactly one primary UniProt keyword; none of those primary keywords were reused as the primary keyword on another ViralZone page. The links box disagreed with the primary keyword on 4 pages, so this report uses the primary keyword as the ViralZone-backed mapping.

In virus.ddb, the SPKW rows carry the exact keyword in with_or_from, such as UniProtKB-KW:KW-1113. This lets us split SPKW annotations into ViralZone-primary KW annotations versus all other UniProt keyword annotations.

For ViralZone-primary keywords with UniProt GO cross-references, the observed virus.ddb SPKW rows matched the UniProt KW->GO mapping exactly: 0 mismatched GO IDs across ViralZone-primary SPKW rows. This supports treating the ViralZone-primary keyword set as a curated vocabulary layer rather than an accidental annotation source.

The important curation consequence is a different prior, not automatic acceptance. ViralZone-primary keyword rows are much less dominated by the broad high-risk terms driving the virus-wide SPKW burden. The broad terms most often flagged in this report are all from non-ViralZone-primary keywords:

Top ViralZone-primary naive SPKW-unique terms show the contrast:

Working rule: mark ViralZone-primary KW rows as VZ-primary in review notes. This should raise the prior that the term is a deliberate viral biology term, especially for entry, tail, genome-ejection, immune-pathway-specific, and restriction-modification terms. It should not override gene-level review: polyprotein granularity, wrong host context, TrEMBL keyword assignment, and multi-GO fanout can still produce over-annotations.

Top Naive SPKW-Unique Terms

Top Viral Taxa by Naive SPKW-Unique Burden

Taxon labels and higher-level clades were resolved from NCBI Taxonomy.

Clade-Specific Patterns

Bacteriophages / Caudoviricetes

Status: multiple reviews completed, including SPKW-BPT4.md, Tequatrovirus genes, anti-CRISPR proteins, phage quorum-sensing, and diversity-generating retroelement proteins.

Main pattern: GO immune and defense terms often encode a eukaryotic-host frame that does not apply to bacterial hosts. Bacterial hosts have restriction-modification, CRISPR-Cas, abortive infection, and related anti-phage systems, not "innate immune response" in the GO immune-system sense.

Representative decisions:

Recommended replacements where supported:

Poxviruses / Pokkesviricetes

Status: not yet reviewed as full gene-review batches in this project, but high-volume candidates are present.

Poxviruses are a different case from phages. "Suppression of host innate immune response" may be biologically real for many poxviral immunomodulators, but the broad parent term can hide whether the protein blocks interferon signaling, NF-kappaB signaling, apoptosis, chemokines, complement, antigen presentation, or host-cell entry.

High-priority terms:

Priority batch: vaccinia and monkeypox immune evasion proteins. Do not blanket-remove the immune term; review mechanism and replace with specific host-pathway child terms when available.

Herpesviruses / Herviviricetes

Status: not yet reviewed as a focused SPKW batch.

Herpesviruses have real latency, host immune evasion, envelope, entry, and nuclear trafficking biology. The main expected error is not semantic impossibility, but over-generalization and missing specificity.

Priority patterns:

Giant dsDNA Viruses / Megaviricetes

Status: high candidate burden; no focused gene-review batch yet.

Mimiviridae and related giant-virus taxa dominate the top taxa table. Their large genomes produce many broad SPKW-derived enzymatic and structural annotations. Many will be directionally correct, but several are probably too general to retain as core function.

Expected patterns:

Priority batch: one Mimiviridae representative with high SPKW burden, preferably Acanthamoeba polyphaga mimivirus or Tupanvirus soda lake.

RNA Viruses

Status: small-proteome viruses are underrepresented in the top virus-wide burden table but have highly specific host-interaction annotations.

The local ZIKV.ddb contains 85 annotations on 3 proteins, including 36 SPKW annotations on 1 protein. Naive SPKW-unique terms include viral entry, viral envelope, host JAK-STAT suppression, host type I interferon suppression, host autophagy modulation, and capping/processing terms.

Key issue: many RNA virus UniProt records are polyproteins. A single polyprotein entry may carry terms that apply to different mature cleavage products. Reviews should be careful not to treat every SPKW-derived term as a function of every mature protein product.

Priority batch: Zika virus polyprotein or mature NS proteins, with special attention to mature-product granularity.

Reviewed Viral SPKW Annotations

Existing viral gene reviews in this repo include 11 genes and 31 SPKW-derived annotations:

If MODIFY, REMOVE, and MARK_AS_OVER_ANNOTATED are counted as annotations requiring curation changes, the current viral pilot issue rate is 17/31 (55%).

Reviewed Genes

Manual Spot-Check Round 1

The virus-wide table above is only a prioritization queue. Following the strategy used in the organism-specific SPKW projects, the next step was to inspect row-level examples by term and clade, then classify whether the SPKW-derived term is likely acceptable, too broad, semantically inverted, or dependent on a specific viral host context.

This first manual pass sampled high-volume terms from virus.ddb on 2026-05-21:

Sample-Level Findings

Term-Level Working Rules

These are not final global edits; they are rules for the next gene-level review batches.

Manual Spot-Check Round 2

The second manual pass expanded coverage beyond the highest-risk terms. It used raw SPKW counts, per-term row samples, and clade-specific profiles for poxviruses, ASFV, herpesviruses, giant dsDNA viruses, RNA viruses, retroviruses, phages, adenoviruses, papillomaviruses, and cyanophage/algal-virus auxiliary metabolism.

Expanded Coverage

Representative Rows Reviewed in Round 2

These rows are examples from the SPKW records inspected for this round. They are not proposed final gene reviews; they show the evidence trail behind the term-level calls.

Manual Spot-Check Round 3: ViralZone-Primary Keywords

This pass focused on the ViralZone-primary subset. The goal was to test whether VZ-primary status behaves like a useful confidence flag and to identify failure modes that remain even when the keyword is ViralZone-backed.

VZ-Primary Working Assessment

This round supports the stratification rule above. The VZ-primary rows sampled here were mostly stronger than generic SPKW rows because they encode a viral mechanism: long-tail ejection, contractile-tail ejection, receptor attachment, restriction-modification evasion, or interferon-pathway inhibition. The remaining risks are narrower:

Manual Spot-Check Round 4: More Granular ViralZone Terms

This pass deliberately sampled lower-level VZ-primary terms where the term name already describes a specific mechanism, substrate, host pathway component, or entry route. Counts are raw SPKW IEA rows in virus.ddb.

Round 4 Granularity Takeaways

Additional Term-Level Findings

Clade Calls After Round 2

High-Priority Virus-Wide Term Batches

Query Used

-- Database: ~/repos/go-db/db/virus.ddb
-- Naive SPKW-unique viral annotations: no non-SPKW evidence for the same gene-term pair.

WITH spkw AS (
    SELECT db_object_id, db_object_symbol, db_object_taxon, ontology_class_ref, aspect
    FROM gaf_association
    WHERE supporting_references = 'GO_REF:0000043'
),
other AS (
    SELECT DISTINCT db_object_id, ontology_class_ref
    FROM gaf_association
    WHERE supporting_references != 'GO_REF:0000043'
),
naive AS (
    SELECT s.*
    FROM spkw s
    WHERE NOT EXISTS (
        SELECT 1
        FROM other o
        WHERE o.db_object_id = s.db_object_id
          AND o.ontology_class_ref = s.ontology_class_ref
    )
)
SELECT
    n.ontology_class_ref,
    tl.label,
    n.aspect,
    COUNT(*) AS annotations,
    COUNT(DISTINCT n.db_object_id) AS proteins,
    COUNT(DISTINCT n.db_object_taxon) AS taxa
FROM naive n
LEFT JOIN term_label tl ON n.ontology_class_ref = tl.id
GROUP BY n.ontology_class_ref, tl.label, n.aspect
ORDER BY annotations DESC;

Closure-filtered SPKW-unique annotations were also computed using isa_partof_closure, matching the method in SPKW-METHODOLOGY.md. For viruses, both views are useful:

• Naive gives the practical review queue, especially for poorly curated viral taxa.
• Closure-filtered reduces redundant broad annotations and is better for prioritizing well-studied viruses with substantial non-SPKW evidence.

Curation Recommendations for Viral SPKW

1. Always check host biology. Phage-bacterium interactions should not inherit eukaryotic immune terms. Use restriction-modification or CRISPR-Cas terms when that is the actual mechanism.

2. Do not blanket-remove immune evasion terms from eukaryotic viruses. Poxviruses, herpesviruses, flaviviruses, and influenza viruses genuinely suppress host innate immune pathways. The right action is often MODIFY to a more specific host-pathway term.

3. Treat broad enzymatic SPKW terms as non-core unless they are the best available term. Viral polymerases, methyltransferases, proteases, and nucleases should usually get specific MF terms.

4. Separate viral life-cycle terms from generic cellular terms. Prefer viral DNA genome replication over generic DNA replication when reviewing viral DNA polymerases and replication proteins.

5. Review component terms at the correct structural level. "Virion component" is often true but rarely sufficient. Use capsid, envelope, virion membrane, tail, nucleocapsid, or other specific CC terms where supported.

6. Handle polyproteins explicitly. For RNA virus polyproteins, record which mature product supports each function and avoid treating all terms as functions of the uncleaved precursor.

7. Use ViralZone-primary keyword status as a confidence flag, not evidence by itself. A VZ-primary SPKW row is less likely to be one of the generic keyword artifacts, but it still needs protein-level review. Non-VZ broad parent terms should be prioritized for cleanup.

Next Review Batches

Relationship to Existing SPKW Projects