SPKW Bacteriophage T4 (BPT4) Subproject

Parent project: SPKW.md

Overview

Bacteriophages present a unique test case for SPKW annotation quality. Many GO terms relating to host-pathogen interactions were designed for eukaryotic pathogens (viruses, bacteria, fungi) infecting eukaryotic hosts. When these terms are applied to bacteriophages (which infect bacteria), semantic mismatches can occur.

Key question: Do GO terms for broad "immune suppression" and "defense response" capture bacteriophage biology, or should they be replaced by precise bacterial anti-phage counter-defense and viral release terms?

Statistics

Updated virus.ddb counts for Enterobacteria phage T4 (taxon:10665) on 2026-05-21:

The original three gene reviews deliberately targeted high-risk SPKW rows, so their 3/3 failure rate should not be read as the rate for all T4 SPKW. The ViralZone split shows a more useful pattern: broad non-VZ keywords drive the worst over-annotations, while many VZ-primary keywords are phage-specific and are often better replacement targets.

ViralZone-Primary Keyword Analysis

For this section, VZ-primary means the UniProt keyword is the primary keyword declared in a ViralZone page header. This is a confidence flag, not evidence by itself. It helps distinguish deliberate viral biology terms from generic UniProt keywords such as hydrolase, virion, DNA binding, or broad immune-response labels.

High-Signal VZ-Primary T4 Rows

VZ-Primary Rows That Still Need Caution

Non-VZ Rows Driving the Original BPT4 Failures

VZ Impact on the BPT4 Conclusion

The BPT4 conclusion should be more nuanced than "SPKW is bad for phages":

1. The original local reviews remain valid: DAM/P04392, E/P00720, and frd/P04382 expose real SPKW over-annotation.
2. ViralZone-primary status strongly improves the prior for phage-specific terms: tail proteins, baseplate proteins, contractile-tail ejection, receptor attachment, peptidoglycan-disruption entry, R-M evasion, DNA-end protection, and CRISPR-cas evasion are mostly valid in the T4 spot check.
3. VZ-primary is not automatic acceptance: agt/P04519 and ndd/P15556 show that broad parent terms can remain too general even when the keyword is ViralZone-backed.
4. For T4 curation, the most actionable cleanup pattern is not "remove all SPKW"; it is "replace broad non-VZ parent terms with the precise VZ-primary phage mechanism where supported."

Full YAML Review Pass

The BPT4 gene-review YAML files were then reviewed directly, not only via row-level spot checks. This pass reviewed every existing_annotations entry in the three local review files and updated action calls so that precise supported functions remain core, while broad parent terms are kept only as non-core.

The only validator warning after this pass is intentional: E/P00720 has three GO:0003796 lysozyme activity entries that remain accepted and one GO:0003796 entry removed because its original reference is wrong. This is a reference-level curation exception, not a disagreement about T4 lysozyme activity.

Deeper Evidence Review

The follow-up pass applied the AIGR review-skill standard explicitly: do not re-check schema details after validation; instead check biological judgment, GO specificity, evidence quality, and core-function coherence. Manual notes were added for each reviewed gene so the evidence trail is visible outside the rendered action table.

Highlighted Gene Deep Research Coverage

All 49 BPT4 genes explicitly highlighted in the VZ/non-VZ tables and case summaries now have provider deep-research files. The three original full-review genes already had Falcon reports; the remaining highlighted genes were fetched with UniProt/GOA context and backfilled with Falcon reports on 2026-05-22. Coverage manifest: reports/spkw_bpt4_highlighted_deep_research_coverage.tsv.

Problematic Term Categories

1. Eukaryotic immune terms applied to phage-bacteria interactions:
2. GO:0052170 (symbiont-mediated suppression of host innate immune response)
3. GO:0052031 (symbiont-mediated perturbation of host defense response)

4. GO:0042742 (defense response to bacterium)

5. Antibiotic response terms applied to phage enzymes:

6. GO:0046677 (response to antibiotic)
7. GO:0031427 (response to methotrexate)

Status

• [x] Initial exploration complete (2026-01-31)
• [x] Deep research for all 3 genes (2026-01-31)
• [x] Annotation review complete for all 3 genes (2026-01-31)
• [x] Write-up complete (2026-01-31)

Case Studies (3 genes reviewed)

Case 1: DAM - DNA Adenine Methyltransferase

Gene: DAM/P04392 - DNA adenine methylase

Review file: genes/BPT4/DAM/DAM-ai-review.yaml

SPKW annotations to review:
- GO:0052170 (symbiont-mediated suppression of host innate immune response)
- GO:0052031 (symbiont-mediated perturbation of host defense response)

Review decision: MODIFY both annotations to GO:0099018 (symbiont-mediated evasion of host restriction-modification system)

Analysis: This is a clear case where a broad immune/defense keyword should be converted to the specific phage counter-defense mechanism:

GO:0052170 Definition:
"Suppression by the symbiont of the innate immune response of the host organism"

The problem:
┌─────────────────────────────────────────────────────────────────────────────┐
│ The evidence does not show general host innate-immune suppression          │
│                                                                             │
│ Eukaryotic innate immunity: Pattern recognition receptors (TLRs), cytokines│
│ Bacterial anti-phage defense: Restriction-modification, CRISPR-Cas, Abi    │
│                                                                             │
│ T4 Dam fits the restriction-modification evasion branch, not a broad parent│
└─────────────────────────────────────────────────────────────────────────────┘

Replacement term (already present): GO:0099018 (symbiont-mediated evasion of host restriction-modification system)
- Definition explicitly mentions phages and bacterial R-M systems
- Accurately describes T4 Dam's biological role

T4 Dam's actual function:
1. Methylates GATC sites on phage DNA (Dam methylation)
2. Protects phage DNA from host restriction enzymes (DpnI, EcoRV, etc.)
3. May also regulate phage gene expression timing

Why the SPKW keyword is too broad: UniProt keyword "Inhibition of host innate immune response by virus" is not the right endpoint for T4 Dam. The supported biology is DNA adenine methylation that lets phage DNA evade host restriction enzymes, so the curation action is a replacement rather than a bare removal.

Case 2: E - T4 Lysozyme

Gene: E/P00720 - T4 phage lysozyme

Review file: genes/BPT4/E/E-ai-review.yaml

SPKW annotations to review:
- GO:0042742 (defense response to bacterium)
- GO:0031640 (killing of cells of another organism)

Review decision: REMOVE GO:0042742; MODIFY GO:0031640 to GO:0044659 (viral release from host cell by cytolysis)

Analysis: This annotation inverts the biological relationship:

GO:0042742 Definition:
"Reactions triggered in response to the presence of a bacterium that act
to protect the cell or organism"

How this term is properly used:
  Eukaryote (human) → produces lysozyme → kills invading bacteria
  This is DEFENSE

How it's misapplied to T4 lysozyme:
  T4 phage → produces lysozyme → lyses host bacterium for viral release
  This is PARASITISM, not defense!

T4 lysozyme's actual function:
1. Hydrolyzes peptidoglycan (beta-1,4 glycosidic bonds between NAM and NAG)
2. Part of the holin-endolysin-spanin lysis system
3. Enables progeny phage release through host cell lysis

Correct annotation: GO:0044659 (viral release from host cell by cytolysis) - already present and correctly describes the biological process.

Additional issue found:
- GO:0031640 (killing of cells of another organism) is directionally related to lysis but too broad.
- The host bacterium is the organism in which the phage replicates; viral release by cytolysis is the defensible process term.

Reference error noted: PMID:4582731 is about T7 lysozyme (an amidase), not T4 lysozyme (a muramidase). The citation was incorrectly associated in the source database.

Case 3: frd - Dihydrofolate Reductase

Gene: frd/P04382 - Dihydrofolate reductase (T4 DHFR)

Review file: genes/BPT4/frd/frd-ai-review.yaml

SPKW annotations to review:
- GO:0046677 (response to antibiotic)
- GO:0031427 (response to methotrexate)

Review decision: REMOVE both annotations

Analysis: This conflates "being a drug target" with "responding to a drug":

The logic error:
┌─────────────────────────────────────────────────────────────────────────────┐
│ UniProt keywords: "Antibiotic resistance", "Trimethoprim resistance"        │
│                              ↓ (mapped to)                                  │
│ GO term: "response to antibiotic"                                           │
│                                                                             │
│ Problem: Being INHIBITED by an antibiotic ≠ RESPONDING to an antibiotic    │
│                                                                             │
│ - DHFR is the TARGET of trimethoprim (inhibits the enzyme)                  │
│ - Phages don't have "antibiotic response pathways"                          │
│ - T4 DHFR does NOT confer trimethoprim resistance                          │
└─────────────────────────────────────────────────────────────────────────────┘

Literature evidence (PMID:32253217, Sanchez-Osuna 2020):

"Phage-encoded DHFRs do NOT confer trimethoprim resistance despite homology"
"Phage folA genes primarily serve phage nucleotide metabolism rather than resistance"

T4 DHFR's actual function:
1. Reduces dihydrofolate to tetrahydrofolate
2. Essential for thymidylate synthesis during phage DNA replication
3. Contributes to the phage's unique nucleotide pool (hydroxymethylcytosine production)

Correct annotations (retained):
- GO:0004146 (dihydrofolate reductase activity)
- GO:0046654 (tetrahydrofolate biosynthetic process)

Reviewed High-Risk Over-Annotation Patterns

Summary: Reviewed High-Risk Cases 3/3 Problematic

All three locally reviewed high-risk T4 phage genes had over-annotated SPKW-unique rows:

Key Lessons from Bacteriophage Analysis

1. GO terms encode eukaryote-centric biology: Terms like "innate immune response" and "defense response" assume eukaryotic biology. Bacterial anti-phage systems (R-M, CRISPR) require different terminology.

2. Host-pathogen terms need taxonomic context: A term describing how a virus evades mammalian immunity should not be applied to a phage evading bacterial restriction.

3. Some UniProt keywords are eukaryote-biased: Keywords like "Inhibition of host innate immune response by virus" were created for eukaryotic viruses and don't translate cleanly to bacteriophages. ViralZone-primary phage terms are less prone to this problem but still need protein-level review.

4. "Defense" depends on perspective: From the phage's perspective, lysozyme enables reproduction. From the bacterium's perspective, it's attack. Neither is "defense response to bacterium."

5. Drug target ≠ drug response: An enzyme being inhibited by an antibiotic doesn't mean the organism "responds" to that antibiotic in a biological process sense.

6. ViralZone helps separate signal from noise: In T4, VZ-primary terms for tail structure, genome ejection, host-envelope penetration, R-M evasion, DNA-end protection, and CRISPR-cas evasion are usually stronger than the broad non-VZ parent terms that caused the original failures.

Recommendations for Phage/Virus SPKW Curation

1. Create phage-specific term mappings: Keywords about host-pathogen interactions should map to different GO terms for phages vs. eukaryotic viruses

2. Review all "innate immune" annotations on phage proteins: These likely need replacement with precise phage counter-defense terms where the mechanism is known

3. Distinguish R-M evasion from immune suppression: GO:0099018 exists specifically for phage-bacteria interactions

4. Validate "antibiotic resistance" claims: Being a drug target does not equal conferring resistance, especially for phage enzymes

5. Flag VZ-primary rows separately: Treat ViralZone-primary status as a positive prior for phage-specific mechanisms, but keep reviewing broad parent terms and protein assignment.

6. Prefer precise phage mechanisms over broad parents: Use R-M evasion, CRISPR-cas evasion, DNA-end degradation evasion, tail fiber/baseplate/sheath/tube terms, contractile-tail ejection, and peptidoglycan-disruption entry when supported.

Comparison: All Organisms Analyzed

Review Files Location

genes/BPT4/
├── DAM/DAM-ai-review.yaml   (MODIFY: broad immune/defense rows -> R-M evasion)
├── E/E-ai-review.yaml       (REMOVE/MODIFY: defense/cell-killing rows -> viral release)
└── frd/frd-ai-review.yaml   (REMOVE: antibiotic response - drug target ≠ response)