GO-DB Query Skill

Overview

This skill provides expertise for querying GO-DB DuckDB databases containing Gene Ontology (GO) annotations. GO-DB databases store annotations linking genes/proteins to GO terms, along with the full GO ontology structure. The key feature is the use of closure tables that enable efficient hierarchical queries across the ontology graph.

Use this skill when working with queries involving:

• Finding genes annotated to specific GO terms (including descendants)
• Analyzing evidence codes and annotation sources
• Exploring ontology hierarchies and term relationships
• Computing annotation statistics by taxon, evidence, or other dimensions
• Identifying unique or redundant annotations using ontological reasoning

Core Concepts

Closure Tables

Closure tables are the heart of GO-DB querying. They contain the transitive closure of ontological relationships:

• isa_partof_closure: Contains all is-a and part-of relationships, both direct and inferred

  • Example: If "protein kinase" is-a "kinase" and "kinase" is-a "catalytic activity", the table includes all three relationships plus the transitive "protein kinase" → "catalytic activity"

• How to use: Join annotations with closure tables to find all genes annotated to a term OR its descendants

-- Find all yeast kinases (including specific types like protein kinase)
SELECT DISTINCT a.db_object_symbol, a.db_object_id
FROM gaf_association a
INNER JOIN isa_partof_closure ipc ON a.ontology_class_ref = ipc.subject
WHERE ipc.object = 'GO:0016301'  -- kinase activity
  AND a.db_object_taxon LIKE '%559292%';  -- yeast

Database Structure

Available databases are located in

db/*.ddb

:

• Organism-specific:

  sgd.ddb

  (yeast),

  fb.ddb

  (fly),

  pombase.ddb

  (fission yeast)
• Taxonomic groups:

  mammal.ddb

  ,

  fungi.ddb

  ,

  plant.ddb

• GOA databases:

  goa_human.ddb

  ,

  goa_uniprot_all.ddb

Query Patterns

1. Finding Genes by GO Term (with Closure)

The most common pattern: find all genes annotated to a term or its descendants.

Pattern:

SELECT DISTINCT
    a.db_object_symbol,
    a.ontology_class_ref,
    t.label AS term_label
FROM gaf_association a
INNER JOIN isa_partof_closure ipc ON a.ontology_class_ref = ipc.subject
INNER JOIN term_label t ON a.ontology_class_ref = t.id
WHERE ipc.object = '<GO_TERM_ID>'
  AND a.db_object_taxon LIKE '%<TAXON_ID>%';

Why use closure: Without the closure join, only direct annotations are found. The closure captures all annotations to descendant terms (e.g., "protein kinase", "lipid kinase" when searching for "kinase").

2. Counting and Grouping Annotations

Aggregate annotations by dimensions like evidence type, taxon, or assigned_by.

Pattern:

SELECT
    evidence_type,
    COUNT(*) AS annotation_count,
    COUNT(DISTINCT db_object_id) AS unique_genes
FROM gaf_association
WHERE <filters>
GROUP BY evidence_type
ORDER BY annotation_count DESC;

Combine with closure tables to count within ontology subtrees.

3. Finding Unique Contributions

Identify annotations that are not redundant with more specific annotations from other sources.

Pattern:

SELECT a.*
FROM gaf_association a
WHERE NOT EXISTS (
    SELECT 1
    FROM gaf_association a2
    INNER JOIN isa_partof_closure ipc ON a2.ontology_class_ref = ipc.subject
    WHERE a2.supporting_references != a.supporting_references
      AND ipc.object = a.ontology_class_ref  -- a2 is to a child term
      AND a2.db_object_id = a.db_object_id
);

Logic: An annotation is unique if no child-term annotation exists from a different source for the same gene.

4. Exploring Term Hierarchies

Navigate the ontology structure itself using edge and closure tables.

Find direct children:

SELECT DISTINCT e.subject, t.label
FROM edge e
INNER JOIN term_label t ON e.subject = t.id
WHERE e.object = '<GO_TERM_ID>'
  AND e.predicate = 'rdfs:subClassOf';

Find all ancestors:

SELECT DISTINCT ipc.object, t.label
FROM isa_partof_closure ipc
INNER JOIN term_label t ON ipc.object = t.id
WHERE ipc.subject = '<GO_TERM_ID>';

5. Genes Annotated to Multiple Terms

Find genes with annotations to both T1 and T2 (or their descendants).

Pattern:

SELECT DISTINCT a1.db_object_symbol, a1.db_object_id
FROM gaf_association a1
INNER JOIN isa_partof_closure ipc1 ON a1.ontology_class_ref = ipc1.subject
INNER JOIN gaf_association a2 ON a1.db_object_id = a2.db_object_id
INNER JOIN isa_partof_closure ipc2 ON a2.ontology_class_ref = ipc2.subject
WHERE ipc1.object = '<GO_TERM_1>'
  AND ipc2.object = '<GO_TERM_2>';

Logic: Self-join gaf_association on gene ID, then join each side with closure tables to check ancestry.

Executing Queries

Command Line Usage

# Query a specific database
duckdb db/sgd.ddb "SELECT COUNT(*) FROM gaf_association"

# Interactive mode
duckdb db/sgd.ddb
D SELECT * FROM term_label WHERE label LIKE '%kinase%' LIMIT 10;
D .quit

# Export results to CSV
duckdb db/goa_human.ddb "COPY (SELECT ...) TO 'results.csv' (HEADER, DELIMITER ',')"

Finding the Right Database

• Organism-specific queries: Use organism database (e.g.,

  sgd.ddb

  for yeast)
• Cross-species analysis: Use taxonomic group (e.g.,

  mammal.ddb

  )
• Human-focused: Use

  goa_human.ddb

• Comprehensive queries: Use

  goa_uniprot_all.ddb

  (largest, >400M annotations)

Check available databases:

ls -lh db/*.ddb

Key Tables Reference

gaf_association

Main annotation table with columns:

• db_object_symbol

  ,

  db_object_id

  : Gene identifier and symbol

• ontology_class_ref

  : GO term ID (e.g., "GO:0016301")

• evidence_type

  : Evidence code (e.g., "IEA", "IDA")

• db_object_taxon

  : NCBI taxon ID (e.g., "taxon:9606")

• aspect

  : GO aspect - "P" (process), "F" (function), "C" (component)

• supporting_references

  : Reference IDs

• assigned_by

  : Annotation source

isa_partof_closure

Transitive closure table with columns:

• subject

  : Descendant term ID

• predicate

  : Relationship type

• object

  : Ancestor term ID

term_label

Term ID to label mapping:

• id

  : GO term ID

• label

  : Human-readable label

entailed_edge

All ontology relationships (including inferred):

• subject

  ,

  predicate

  ,

  object

For complete schema documentation, refer to

references/schema.md

.

Query Workflow

When handling a query request:

1. Understand the question: Identify what data is being requested
2. Determine if closure is needed: Most queries benefit from closure tables to capture hierarchical relationships
3. Find the GO term ID: Use term_label to search by label if needed
4. Select the right database: Choose based on organism/scope
5. Build the query: Start with the appropriate pattern from

   references/common_queries.md

6. Add filters: Refine by taxon, evidence, date, etc.
7. Execute and verify: Run via

   duckdb

   and check results make sense
8. Add labels for readability: Join with term_label to show human-readable names

Common Taxon IDs

• 9606: Human
• 10090: Mouse
• 559292: S. cerevisiae (yeast)
• 7227: D. melanogaster (fly)
• 284812: S. pombe (fission yeast)

Common Evidence Codes

Experimental: IDA, IMP, IGI, IPI, IEP Computational: IEA, ISS, ISO, ISA, ISM, IBA Curator/Author: TAS, NAS, IC, ND

Resources

references/schema.md

Complete schema documentation including:

• Detailed table structures and column descriptions
• Ontology table relationships
• Closure table explanations
• Index information
• Database statistics

references/common_queries.md

Comprehensive SQL examples for all query patterns:

• Pattern 1: Find genes by term (with closure)
• Pattern 2: Count/group annotations
• Pattern 3: Find unique/redundant annotations
• Pattern 4: Explore term hierarchies
• Pattern 5: Genes with multiple term annotations
• Pattern 6: Evidence analysis
• Pattern 7: Reference/citation analysis

Load these references when detailed examples or schema information is needed to construct queries.

Tips

• Start simple: Begin with basic queries and add complexity incrementally
• Use EXPLAIN: Check query plans for complex queries
• LIMIT during development: Add LIMIT to test queries on large databases
• Check indices: Closure tables have indices on subject/object pairs for performance
• Validate term IDs: Verify GO term IDs exist in term_label before running queries
• Consider performance: Closure joins can be expensive on very large databases; filter early when possible