tools-only Interaction Databases - Usage Guide
Overview
This skill enables AI agents to query protein-protein interaction (PPI) databases including STRING, BioGRID, IntAct, and OmniPath. Retrieves interaction partners, confidence scores, and functional enrichment for gene sets, then converts results into NetworkX graphs for downstream analysis.
Prerequisites
pip install requests pandas networkx
Optional for specific databases:
# OmniPath Python client (alternative to REST)
pip install omnipath
# STRINGdb R package (alternative to REST)
# install.packages('BiocManager')
# BiocManager::install('STRINGdb')
- BioGRID requires a free API key from thebiogrid.org
- STRING, IntAct, and OmniPath are key-free
Quick Start
Tell your AI agent what you want to do:
- "Get all protein interactions for TP53 from STRING"
- "Build a PPI network for my list of DE genes"
- "Find high-confidence interactions between these kinases"
- "Query BioGRID for physical interactions with BRCA1"
- "Combine STRING and OmniPath interactions into one network"
Example Prompts
STRING Queries
"Get STRING interactions for TP53, MDM2, BRCA1, ATM, and CHEK2 with high confidence"
"Download the STRING network image for my DNA damage response genes"
"Run STRING enrichment analysis on my upregulated gene list"
BioGRID Queries
"Find all low-throughput physical interactions for MYC in BioGRID"
"Get BioGRID interactions for EGFR filtered to co-immunoprecipitation experiments"
Multi-Database
"Query both STRING and OmniPath for my gene list and merge the results"
"Build a combined PPI network from STRING, BioGRID, and IntAct for these 50 genes"
Network Construction
"Convert my STRING interactions to a NetworkX graph and find the hub genes"
"Build a PPI network and compute centrality measures for my gene list"
What the Agent Will Do
- Resolve gene identifiers to database-specific IDs
- Query one or more interaction databases via REST APIs
- Filter interactions by confidence score and evidence type
- Convert results to a NetworkX graph
- Compute network statistics (degree, clustering, components)
- Export the network for visualization or downstream analysis
Tips
- Score threshold 700 - Use high-confidence STRING scores (700+) for publication networks; use 400 for exploratory analysis where recall matters
- BioGRID API key - Free but required; register at thebiogrid.org for access
- Low-throughput evidence - Filter BioGRID to low-throughput experiments for higher-quality physical interactions
- OmniPath for signaling - OmniPath curates directed signaling interactions, ideal for pathway reconstruction
- Multi-database consensus - Interactions found in multiple databases are more reliable; aggregate and filter by source count
- Species codes - STRING uses NCBI taxonomy IDs: 9606 (human), 10090 (mouse), 7955 (zebrafish), 7227 (fly)
- Rate limiting - Add delays between batch queries to STRING and BioGRID; process gene lists in chunks of 50-100
Related Skills
- database-access/uniprot-access - Protein annotations and cross-references
- pathway-analysis/go-enrichment - Functional enrichment of network genes
- gene-regulatory-networks/coexpression-networks - Co-expression network construction
- data-visualization/network-visualization - Visualize interaction networks