View original ARBA rule on UniProt
Catastrophically broad rule applying the "Transferase" keyword (KW-0808) to over 14 million proteins using 4,473 condition sets encompassing 1,932 unique InterPro domains and 3,730 CATH FunFam families. This represents the most extreme case of over-annotation in the ARBA system, conflating virtually all transferase activities under a single meaningless keyword.
Condition-set counts describe the sets recorded in this review, which may omit the full rule.
This rule is a prime example of destructive over-annotation that should be immediately REMOVED. It applies the broad 'Transferase' keyword to over 14 million proteins based on the presence of any of thousands of different InterPro domains or FunFam families. This approach violates fundamental principles of functional annotation specificity and creates massive noise in UniProt annotations. The rule conflates diverse enzymatic activities - kinases, methyltransferases, acetyltransferases, aminotransferases, glycosyltransferases, and many others - under a single meaningless umbrella term. This type of annotation provides no useful biological insight and actively degrades annotation quality.
This rule should be removed entirely for multiple critical reasons: (1) Massive over-annotation - 14+ million proteins is far too broad for any single functional annotation, (2) Biological meaninglessness - the 'Transferase' keyword is too general to provide useful functional information, (3) Conflation of distinct enzyme classes - combining kinases, methyltransferases, phosphatases, and other unrelated activities under one term, (4) Lack of specificity - any protein with any transferase-related domain gets the same annotation regardless of actual activity, (5) Noise generation - this rule likely creates more false positives than true functional annotations. The rule represents exactly the type of broad, non-specific annotation that the GO Consortium and UniProt community have moved away from in favor of precise, evidence-based functional annotation.
This rule contains 4,473 condition sets, each typically containing a single InterPro domain or CATH FunFam family. The 1,932 unique InterPro domains span virtually every transferase family including protein kinases, methyltransferases, acetyltransferases, aminotransferases, glycosyltransferases, phosphatases, and hundreds of other enzyme classes. The additional 3,730 FunFam families further expand coverage to include structural homologs. This represents a catastrophic failure of annotation specificity.
This rule represents the absolute antithesis of parsimony. With 4,473 condition sets targeting disparate enzyme families across 1,932 InterPro domains and 3,730 FunFam families, it violates every principle of specific, targeted annotation. A parsimonious approach would create hundreds of separate rules for individual enzyme families (e.g., serine/threonine kinases, SAM-dependent methyltransferases) with appropriate domain architecture and taxonomic constraints. This rule's extreme complexity stems from the fundamentally flawed approach of applying a single broad keyword to virtually all transferase activities in biology.
While individual enzyme families have strong literature support, the concept of annotating all transferases with a single broad keyword contradicts decades of functional annotation best practices. The literature strongly supports specific, mechanistically-informed annotations rather than broad categorical labels. The EC classification system deliberately separates transferases into distinct subclasses (EC 2.1-2.9) based on chemical mechanism, not broad category. This rule's approach is fundamentally at odds with the precision required for useful functional annotation.
With 4,473 condition sets covering virtually every transferase family known to biology, there is massive conceptual overlap but minimal molecular specificity. The rule inappropriately conflates functionally distinct enzyme classes - protein kinases, DNA methyltransferases, glycosyltransferases, aminotransferases - that share no evolutionary relationship or mechanistic similarity beyond the broad concept of 'catalyzing transfer reactions'. Many condition sets likely have zero protein overlap (e.g., plant-specific vs animal-specific enzymes), while others may have complete redundancy. This represents the ultimate failure of condition set design.
The rule doesn't apply GO terms but rather a UniProt keyword. However, the 'Transferase' keyword is far too broad to be useful. If GO terms were applied, they would need to be specific molecular function terms (e.g., GO:0016740 transferase activity) with appropriate child terms for each enzyme class rather than a single broad parent term.
The rule has no taxonomic restrictions, applying the transferase keyword across all domains of life. While transferases exist in all organisms, the specific families and their functions vary dramatically between taxa. The rule would benefit from taxonomic specificity to account for lineage-specific enzyme evolution and function.
Protein that belongs to the transferase family. Transferases catalyze the transfer of a group, for example a methyl group, glycosyl group, amino group, acyl group, phosphorus-containing group, sulfur-containing group, etc., from one compound (donor) to another compound (acceptor).
Annotations should be as specific as possible. Avoid annotating to high-level terms when more specific terms are available.
This rule contains 4,473 condition sets encompassing 1,932 unique InterPro domains and 3,730 CATH FunFam families, representing virtually every transferase enzyme family known to biology. Examples include completely unrelated enzyme families like histone deacetylases, protein kinases, and metabolic enzymes.
The EC classification system deliberately separates transferases into distinct subclasses (EC 2.1-2.9) based on chemical mechanism. Modern annotation principles emphasize specificity over breadth and mechanistic basis over broad categories.
This rule violates every principle of successful ARBA rule design: it targets millions of proteins instead of hundreds, uses thousands of condition sets instead of fewer than ten, and applies broad keywords instead of specific GO terms.
id: ARBA00022679
description: Catastrophically broad rule applying the "Transferase" keyword (KW-0808) to over 14 million proteins using 4,473 condition sets encompassing 1,932 unique InterPro domains and 3,730 CATH FunFam families. This represents the most extreme case of over-annotation in the ARBA system, conflating virtually all transferase activities under a single meaningless keyword.
status: COMPLETE
rule_type: ARBA
rule:
rule_id: ARBA00022679
condition_sets:
- number: 1
conditions:
- condition_type: INTERPRO
value: IPR001986
curie: InterPro:IPR001986
label: Fumarylacetoacetate hydrolase
negated: false
notes: "This rule contains 4,473 condition sets, each typically containing a single InterPro domain or CATH FunFam family. The 1,932 unique InterPro domains span virtually every transferase family including protein kinases, methyltransferases, acetyltransferases, aminotransferases, glycosyltransferases, phosphatases, and hundreds of other enzyme classes. The additional 3,730 FunFam families further expand coverage to include structural homologs. This represents a catastrophic failure of annotation specificity."
# NOTE: Only showing first condition set - rule actually has 4,473 condition sets
keyword_annotations:
- keyword_id: KW-0808
keyword_label: Transferase
category: Molecular function
reviewed_protein_count: 0
unreviewed_protein_count: 14017890
created_date: '2020-05-12'
modified_date: '2025-05-15'
review_summary: "This rule is a prime example of destructive over-annotation that should be immediately REMOVED. It applies the broad 'Transferase' keyword to over 14 million proteins based on the presence of any of thousands of different InterPro domains or FunFam families. This approach violates fundamental principles of functional annotation specificity and creates massive noise in UniProt annotations. The rule conflates diverse enzymatic activities - kinases, methyltransferases, acetyltransferases, aminotransferases, glycosyltransferases, and many others - under a single meaningless umbrella term. This type of annotation provides no useful biological insight and actively degrades annotation quality."
action: REMOVE
action_rationale: "This rule should be removed entirely for multiple critical reasons: (1) Massive over-annotation - 14+ million proteins is far too broad for any single functional annotation, (2) Biological meaninglessness - the 'Transferase' keyword is too general to provide useful functional information, (3) Conflation of distinct enzyme classes - combining kinases, methyltransferases, phosphatases, and other unrelated activities under one term, (4) Lack of specificity - any protein with any transferase-related domain gets the same annotation regardless of actual activity, (5) Noise generation - this rule likely creates more false positives than true functional annotations. The rule represents exactly the type of broad, non-specific annotation that the GO Consortium and UniProt community have moved away from in favor of precise, evidence-based functional annotation."
suggested_modifications:
- "Remove the entire rule - no salvageable components"
- "Replace with specific, evidence-based rules for individual enzyme families"
- "Use precise GO molecular function terms instead of broad keywords"
- "Implement taxonomic and domain architecture constraints for specificity"
parsimony:
assessment: OVERLY_COMPLEX
notes: "This rule represents the absolute antithesis of parsimony. With 4,473 condition sets targeting disparate enzyme families across 1,932 InterPro domains and 3,730 FunFam families, it violates every principle of specific, targeted annotation. A parsimonious approach would create hundreds of separate rules for individual enzyme families (e.g., serine/threonine kinases, SAM-dependent methyltransferases) with appropriate domain architecture and taxonomic constraints. This rule's extreme complexity stems from the fundamentally flawed approach of applying a single broad keyword to virtually all transferase activities in biology."
literature_support:
assessment: CONTRADICTED
notes: "While individual enzyme families have strong literature support, the concept of annotating all transferases with a single broad keyword contradicts decades of functional annotation best practices. The literature strongly supports specific, mechanistically-informed annotations rather than broad categorical labels. The EC classification system deliberately separates transferases into distinct subclasses (EC 2.1-2.9) based on chemical mechanism, not broad category. This rule's approach is fundamentally at odds with the precision required for useful functional annotation."
supported_by:
- reference_id: "GO_ANNOTATION_PRINCIPLES"
supporting_text: "GO annotations should be as specific as possible while still being correct"
- reference_id: "file:rules/arba/ARBA00022679/ARBA00022679-deep-research-analysis.md"
supporting_text: "The EC classification system divides transferases into numerous subclasses (EC 2.x.x.x), each with distinct biochemical mechanisms. Each subclass represents fundamentally different chemical mechanisms and biological roles."
condition_overlap:
assessment: COMPLETE
notes: "With 4,473 condition sets covering virtually every transferase family known to biology, there is massive conceptual overlap but minimal molecular specificity. The rule inappropriately conflates functionally distinct enzyme classes - protein kinases, DNA methyltransferases, glycosyltransferases, aminotransferases - that share no evolutionary relationship or mechanistic similarity beyond the broad concept of 'catalyzing transfer reactions'. Many condition sets likely have zero protein overlap (e.g., plant-specific vs animal-specific enzymes), while others may have complete redundancy. This represents the ultimate failure of condition set design."
go_specificity:
assessment: TOO_BROAD
notes: "The rule doesn't apply GO terms but rather a UniProt keyword. However, the 'Transferase' keyword is far too broad to be useful. If GO terms were applied, they would need to be specific molecular function terms (e.g., GO:0016740 transferase activity) with appropriate child terms for each enzyme class rather than a single broad parent term."
taxonomic_scope:
assessment: TOO_BROAD
notes: "The rule has no taxonomic restrictions, applying the transferase keyword across all domains of life. While transferases exist in all organisms, the specific families and their functions vary dramatically between taxa. The rule would benefit from taxonomic specificity to account for lineage-specific enzyme evolution and function."
confidence: 0.95
references:
- id: "UniProt_Keyword_KW-0808"
title: "UniProt Keyword: Transferase"
findings:
- statement: "Protein that belongs to the transferase family. Transferases catalyze the transfer of a group, for example a methyl group, glycosyl group, amino group, acyl group, phosphorus-containing group, sulfur-containing group, etc., from one compound (donor) to another compound (acceptor)."
- id: "GO_ANNOTATION_GUIDELINES"
title: "Gene Ontology Annotation Guidelines"
findings:
- statement: "Annotations should be as specific as possible. Avoid annotating to high-level terms when more specific terms are available."
- id: "file:rules/arba/ARBA00022679/ARBA00022679-notes.md"
title: "Quantitative analysis of rule structure and complexity"
findings:
- statement: "This rule contains 4,473 condition sets encompassing 1,932 unique InterPro domains and 3,730 CATH FunFam families, representing virtually every transferase enzyme family known to biology. Examples include completely unrelated enzyme families like histone deacetylases, protein kinases, and metabolic enzymes."
- id: "file:rules/arba/ARBA00022679/ARBA00022679-deep-research-analysis.md"
title: "Literature analysis of transferase classification and annotation principles"
findings:
- statement: "The EC classification system deliberately separates transferases into distinct subclasses (EC 2.1-2.9) based on chemical mechanism. Modern annotation principles emphasize specificity over breadth and mechanistic basis over broad categories."
- statement: "This rule violates every principle of successful ARBA rule design: it targets millions of proteins instead of hundreds, uses thousands of condition sets instead of fewer than ten, and applies broad keywords instead of specific GO terms."
supported_by:
- reference_id: "GO_ANNOTATION_GUIDELINES"
supporting_text: "The goal is to make annotations that are as specific as possible, while still being correct"
- reference_id: "file:rules/arba/ARBA00022679/ARBA00022679-deep-research-analysis.md"
supporting_text: "ARBA00022679 represents everything wrong with broad, automated annotation approaches. It prioritizes quantity over quality, breadth over specificity, and automation over accuracy."