claude-scientific-skills/scientific-databases/string-database/SKILL.md

name, description

name

description

string-database

Work with the STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) database for protein-protein interaction network analysis, functional enrichment, and interaction partner discovery. Use this skill when analyzing protein interactions, building protein networks, performing pathway enrichment on protein sets, finding interaction partners, testing network connectivity significance, or retrieving protein homology data. Covers 5000+ genomes with 59.3 million proteins and 20+ billion interactions from experimental data, computational prediction, and text-mining.

STRING Database

Overview

STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) is a comprehensive database of known and predicted protein-protein interactions. This skill enables interaction with STRING's REST API to query protein networks, analyze functional enrichments, and discover interaction partners across 5000+ organisms.

Database Statistics:

• Coverage: 5000+ genomes
• Proteins: ~59.3 million
• Interactions: 20+ billion
• Evidence types: Experimental data, computational prediction, text-mining, pathway databases

Data Sources: Integrates over 40 sources including experimental repositories, pathway databases, automated text-mining, and computational predictions.

Quick Start

The skill provides:

1. Python helper functions (scripts/string_api.py) for all STRING REST API operations
2. Comprehensive reference documentation (references/string_reference.md) with detailed API specifications

When users request STRING data, determine which operation is needed and use the appropriate function from scripts/string_api.py.

Core Operations

1. Identifier Mapping (string_map_ids)

Convert gene names, protein names, and external IDs to STRING identifiers.

When to use: Starting any STRING analysis, validating protein names, finding canonical identifiers.

Usage:

from scripts.string_api import string_map_ids

# Map single protein
result = string_map_ids('TP53', species=9606)

# Map multiple proteins
result = string_map_ids(['TP53', 'BRCA1', 'EGFR', 'MDM2'], species=9606)

# Map with multiple matches per query
result = string_map_ids('p53', species=9606, limit=5)

Parameters:

• species: NCBI taxon ID (9606 = human, 10090 = mouse, 7227 = fly)
• limit: Number of matches per identifier (default: 1)
• echo_query: Include query term in output (default: 1)

Best practice: Always map identifiers first for faster subsequent queries.

2. Network Retrieval (string_network)

Get protein-protein interaction network data in tabular format.

When to use: Building interaction networks, analyzing connectivity, retrieving interaction evidence.

Usage:

from scripts.string_api import string_network

# Get network for single protein
network = string_network('9606.ENSP00000269305', species=9606)

# Get network with multiple proteins
proteins = ['9606.ENSP00000269305', '9606.ENSP00000275493']
network = string_network(proteins, required_score=700)

# Expand network with additional interactors
network = string_network('TP53', species=9606, add_nodes=10, required_score=400)

# Physical interactions only
network = string_network('TP53', species=9606, network_type='physical')

Parameters:

• required_score: Confidence threshold (0-1000)
  • 150: low confidence (exploratory)
  • 400: medium confidence (default, standard analysis)
  • 700: high confidence (conservative)
  • 900: highest confidence (very stringent)
• network_type: 'functional' (all evidence, default) or 'physical' (direct binding only)
• add_nodes: Add N most connected proteins (0-10)

Output columns: Interaction pairs, confidence scores, and individual evidence scores (neighborhood, fusion, coexpression, experimental, database, text-mining).

3. Network Visualization (string_network_image)

Generate network visualization as PNG image.

When to use: Creating figures, visual exploration, presentations.

Usage:

from scripts.string_api import string_network_image

# Get network image
proteins = ['TP53', 'MDM2', 'ATM', 'CHEK2', 'BRCA1']
img_data = string_network_image(proteins, species=9606, required_score=700)

# Save image
with open('network.png', 'wb') as f:
    f.write(img_data)

# Evidence-colored network
img = string_network_image(proteins, species=9606, network_flavor='evidence')

# Confidence-based visualization
img = string_network_image(proteins, species=9606, network_flavor='confidence')

# Actions network (activation/inhibition)
img = string_network_image(proteins, species=9606, network_flavor='actions')

Network flavors:

• 'evidence': Colored lines show evidence types (default)
• 'confidence': Line thickness represents confidence
• 'actions': Shows activating/inhibiting relationships

4. Interaction Partners (string_interaction_partners)

Find all proteins that interact with given protein(s).

When to use: Discovering novel interactions, finding hub proteins, expanding networks.

Usage:

from scripts.string_api import string_interaction_partners

# Get top 10 interactors of TP53
partners = string_interaction_partners('TP53', species=9606, limit=10)

# Get high-confidence interactors
partners = string_interaction_partners('TP53', species=9606,
                                      limit=20, required_score=700)

# Find interactors for multiple proteins
partners = string_interaction_partners(['TP53', 'MDM2'],
                                      species=9606, limit=15)

Parameters:

• limit: Maximum number of partners to return (default: 10)
• required_score: Confidence threshold (0-1000)

Use cases:

• Hub protein identification
• Network expansion from seed proteins
• Discovering indirect connections

5. Functional Enrichment (string_enrichment)

Perform enrichment analysis across Gene Ontology, KEGG pathways, Pfam domains, and more.

When to use: Interpreting protein lists, pathway analysis, functional characterization, understanding biological processes.

Usage:

from scripts.string_enrichment import string_enrichment

# Enrichment for a protein list
proteins = ['TP53', 'MDM2', 'ATM', 'CHEK2', 'BRCA1', 'ATR', 'TP73']
enrichment = string_enrichment(proteins, species=9606)

# Parse results to find significant terms
import pandas as pd
df = pd.read_csv(io.StringIO(enrichment), sep='\t')
significant = df[df['fdr'] < 0.05]

Enrichment categories:

• Gene Ontology: Biological Process, Molecular Function, Cellular Component
• KEGG Pathways: Metabolic and signaling pathways
• Pfam: Protein domains
• InterPro: Protein families and domains
• SMART: Domain architecture
• UniProt Keywords: Curated functional keywords

Output columns:

• category: Annotation database (e.g., "KEGG Pathways", "GO Biological Process")
• term: Term identifier
• description: Human-readable term description
• number_of_genes: Input proteins with this annotation
• p_value: Uncorrected enrichment p-value
• fdr: False discovery rate (corrected p-value)

Statistical method: Fisher's exact test with Benjamini-Hochberg FDR correction.

Interpretation: FDR < 0.05 indicates statistically significant enrichment.

6. PPI Enrichment (string_ppi_enrichment)

Test if a protein network has significantly more interactions than expected by chance.

When to use: Validating if proteins form functional module, testing network connectivity.

Usage:

from scripts.string_api import string_ppi_enrichment
import json

# Test network connectivity
proteins = ['TP53', 'MDM2', 'ATM', 'CHEK2', 'BRCA1']
result = string_ppi_enrichment(proteins, species=9606, required_score=400)

# Parse JSON result
data = json.loads(result)
print(f"Observed edges: {data['number_of_edges']}")
print(f"Expected edges: {data['expected_number_of_edges']}")
print(f"P-value: {data['p_value']}")

Output fields:

• number_of_nodes: Proteins in network
• number_of_edges: Observed interactions
• expected_number_of_edges: Expected in random network
• p_value: Statistical significance

Interpretation:

• p-value < 0.05: Network is significantly enriched (proteins likely form functional module)
• p-value ≥ 0.05: No significant enrichment (proteins may be unrelated)

7. Homology Scores (string_homology)

Retrieve protein similarity and homology information.

When to use: Identifying protein families, paralog analysis, cross-species comparisons.

Usage:

from scripts.string_api import string_homology

# Get homology between proteins
proteins = ['TP53', 'TP63', 'TP73']  # p53 family
homology = string_homology(proteins, species=9606)

Use cases:

• Protein family identification
• Paralog discovery
• Evolutionary analysis

8. Version Information (string_version)

Get current STRING database version.

When to use: Ensuring reproducibility, documenting methods.

Usage:

from scripts.string_api import string_version

version = string_version()
print(f"STRING version: {version}")

Common Analysis Workflows

Workflow 1: Protein List Analysis (Standard Workflow)

Use case: Analyze a list of proteins from experiment (e.g., differential expression, proteomics).

from scripts.string_api import (string_map_ids, string_network,
                                string_enrichment, string_ppi_enrichment,
                                string_network_image)

# Step 1: Map gene names to STRING IDs
gene_list = ['TP53', 'BRCA1', 'ATM', 'CHEK2', 'MDM2', 'ATR', 'BRCA2']
mapping = string_map_ids(gene_list, species=9606)

# Step 2: Get interaction network
network = string_network(gene_list, species=9606, required_score=400)

# Step 3: Test if network is enriched
ppi_result = string_ppi_enrichment(gene_list, species=9606)

# Step 4: Perform functional enrichment
enrichment = string_enrichment(gene_list, species=9606)

# Step 5: Generate network visualization
img = string_network_image(gene_list, species=9606,
                          network_flavor='evidence', required_score=400)
with open('protein_network.png', 'wb') as f:
    f.write(img)

# Step 6: Parse and interpret results

Workflow 2: Single Protein Investigation

Use case: Deep dive into one protein's interactions and partners.

from scripts.string_api import (string_map_ids, string_interaction_partners,
                                string_network_image)

# Step 1: Map protein name
protein = 'TP53'
mapping = string_map_ids(protein, species=9606)

# Step 2: Get all interaction partners
partners = string_interaction_partners(protein, species=9606,
                                      limit=20, required_score=700)

# Step 3: Visualize expanded network
img = string_network_image(protein, species=9606, add_nodes=15,
                          network_flavor='confidence', required_score=700)
with open('tp53_network.png', 'wb') as f:
    f.write(img)

Workflow 3: Pathway-Centric Analysis

Use case: Identify and visualize proteins in a specific biological pathway.

from scripts.string_api import string_enrichment, string_network

# Step 1: Start with known pathway proteins
dna_repair_proteins = ['TP53', 'ATM', 'ATR', 'CHEK1', 'CHEK2',
                       'BRCA1', 'BRCA2', 'RAD51', 'XRCC1']

# Step 2: Get network
network = string_network(dna_repair_proteins, species=9606,
                        required_score=700, add_nodes=5)

# Step 3: Enrichment to confirm pathway annotation
enrichment = string_enrichment(dna_repair_proteins, species=9606)

# Step 4: Parse enrichment for DNA repair pathways
import pandas as pd
import io
df = pd.read_csv(io.StringIO(enrichment), sep='\t')
dna_repair = df[df['description'].str.contains('DNA repair', case=False)]

Workflow 4: Cross-Species Analysis

Use case: Compare protein interactions across different organisms.

from scripts.string_api import string_network

# Human network
human_network = string_network('TP53', species=9606, required_score=700)

# Mouse network
mouse_network = string_network('Trp53', species=10090, required_score=700)

# Yeast network (if ortholog exists)
yeast_network = string_network('gene_name', species=4932, required_score=700)

Workflow 5: Network Expansion and Discovery

Use case: Start with seed proteins and discover connected functional modules.

from scripts.string_api import (string_interaction_partners, string_network,
                                string_enrichment)

# Step 1: Start with seed protein(s)
seed_proteins = ['TP53']

# Step 2: Get first-degree interactors
partners = string_interaction_partners(seed_proteins, species=9606,
                                      limit=30, required_score=700)

# Step 3: Parse partners to get protein list
import pandas as pd
import io
df = pd.read_csv(io.StringIO(partners), sep='\t')
all_proteins = list(set(df['preferredName_A'].tolist() +
                       df['preferredName_B'].tolist()))

# Step 4: Perform enrichment on expanded network
enrichment = string_enrichment(all_proteins[:50], species=9606)

# Step 5: Filter for interesting functional modules
enrichment_df = pd.read_csv(io.StringIO(enrichment), sep='\t')
modules = enrichment_df[enrichment_df['fdr'] < 0.001]

Common Species

When specifying species, use NCBI taxon IDs:

Organism	Common Name	Taxon ID
Homo sapiens	Human	9606
Mus musculus	Mouse	10090
Rattus norvegicus	Rat	10116
Drosophila melanogaster	Fruit fly	7227
Caenorhabditis elegans	C. elegans	6239
Saccharomyces cerevisiae	Yeast	4932
Arabidopsis thaliana	Thale cress	3702
Escherichia coli	E. coli	511145
Danio rerio	Zebrafish	7955

Full list available at: https://string-db.org/cgi/input?input_page_active_form=organisms

Understanding Confidence Scores

STRING provides combined confidence scores (0-1000) integrating multiple evidence types:

Evidence Channels

1. Neighborhood (nscore): Conserved genomic neighborhood across species
2. Fusion (fscore): Gene fusion events
3. Phylogenetic Profile (pscore): Co-occurrence patterns across species
4. Coexpression (ascore): Correlated RNA expression
5. Experimental (escore): Biochemical and genetic experiments
6. Database (dscore): Curated pathway and complex databases
7. Text-mining (tscore): Literature co-occurrence and NLP extraction

Recommended Thresholds

Choose threshold based on analysis goals:

• 150 (low confidence): Exploratory analysis, hypothesis generation
• 400 (medium confidence): Standard analysis, balanced sensitivity/specificity
• 700 (high confidence): Conservative analysis, high-confidence interactions
• 900 (highest confidence): Very stringent, experimental evidence preferred

Trade-offs:

• Lower thresholds: More interactions (higher recall, more false positives)
• Higher thresholds: Fewer interactions (higher precision, more false negatives)

Network Types

Functional Networks (Default)

Includes all evidence types (experimental, computational, text-mining). Represents proteins that are functionally associated, even without direct physical binding.

When to use:

• Pathway analysis
• Functional enrichment studies
• Systems biology
• Most general analyses

Physical Networks

Only includes evidence for direct physical binding (experimental data and database annotations for physical interactions).

When to use:

• Structural biology studies
• Protein complex analysis
• Direct binding validation
• When physical contact is required

API Best Practices

1. Always map identifiers first: Use string_map_ids() before other operations for faster queries
2. Use STRING IDs when possible: Use format 9606.ENSP00000269305 instead of gene names
3. Specify species for networks >10 proteins: Required for accurate results
4. Respect rate limits: Wait 1 second between API calls
5. Use versioned URLs for reproducibility: Available in reference documentation
6. Handle errors gracefully: Check for "Error:" prefix in returned strings
7. Choose appropriate confidence thresholds: Match threshold to analysis goals

Detailed Reference

For comprehensive API documentation, complete parameter lists, output formats, and advanced usage, refer to references/string_reference.md. This includes:

• Complete API endpoint specifications
• All supported output formats (TSV, JSON, XML, PSI-MI)
• Advanced features (bulk upload, values/ranks enrichment)
• Error handling and troubleshooting
• Integration with other tools (Cytoscape, R, Python libraries)
• Data license and citation information

Troubleshooting

No proteins found:

• Verify species parameter matches identifiers
• Try mapping identifiers first with string_map_ids()
• Check for typos in protein names

Empty network results:

• Lower confidence threshold (required_score)
• Check if proteins actually interact
• Verify species is correct

Timeout or slow queries:

• Reduce number of input proteins
• Use STRING IDs instead of gene names
• Split large queries into batches

"Species required" error:

• Add species parameter for networks with >10 proteins
• Always include species for consistency

Results look unexpected:

• Check STRING version with string_version()
• Verify network_type is appropriate (functional vs physical)
• Review confidence threshold selection

Additional Resources

For proteome-scale analysis or complete species network upload:

• Visit https://string-db.org
• Use "Upload proteome" feature
• STRING will generate complete interaction network and predict functions

For bulk downloads of complete datasets:

• Download page: https://string-db.org/cgi/download
• Includes complete interaction files, protein annotations, and pathway mappings

Data License

STRING data is freely available under Creative Commons BY 4.0 license:

• Free for academic and commercial use
• Attribution required when publishing
• Cite latest STRING publication

Citation

When using STRING in publications, cite the most recent publication from: https://string-db.org/cgi/about