Skip to content
pmallPublic

About

AI-native knowledge graph of virus–human protein–protein interactions.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

bpgraph

A FalkorDB knowledge graph of protein–protein interactions, human–human and virus–human, built to be explored by an agent. Scripts do the deterministic work; the agent reads the connected text — function descriptions, GO annotations, abstracts — and draws the connections a person would need years of reading to make.

The data

The human interactome is IntAct's interactions, kept by detection method, with our own curation on top; the virus–human interactions are ours. Every interaction carries the publications, detection methods and peptides behind it, abstracts included. Every Swiss-Prot human protein is in, with its UniProt function text and experimental GO annotations of what it does — biological process and molecular function — each tied to the publications behind it. A viral protein is a curated one — HBx of HBV, NS5A of HCV — pooled over every strain and UniProt accession it was observed on, so its evidence is counted once rather than scattered; its virus is the species curation groups it under, rolled up to its NCBI family. Curated topics — lists of human proteins involved in a subject such as ferroptosis, which a query takes as a parameter — make it possible to ask how viral families act on a biological process. Sequences live in SQLite vaults beside the graph.

Curation itself happens in a separate relational database. The graph is never edited in place: each run is built from that database's export and from public sources into a staging graph, checked, and swapped in whole. graph/docs/schema.md is what the graph holds, and graph/docs/roadmap.md where it is heading: many hosts, sequence and full-text vaults, protein language model features.

The parts

This repository is a uv workspace. Its package graph/, bpgraph, is the graph and everything around it: the ingestion that builds it, an internal query API over the graph and its vaults, and the MCP server, the only thing consumers see. A repository consulting the graph keeps its own instructions and analysis skills, and reaches the graph through the MCP server only.

Running it

Requires Docker and uv.

cp .env.example .env        # set FALKORDB_PASSWORD to enable auth
docker compose up -d
uv sync

This brings up the graph, its browser UI at http://localhost:3000 — log in with host falkordb-server, not localhost — and the MCP server at http://localhost:8080/mcp; docker-compose.yml describes each service. graph/docs/build.md says how to build a graph, and the MCP server tells a consulting agent how to query it.

AGENTS.md is the guide for agents working on this repository.

About

AI-native knowledge graph of virus–human protein–protein interactions.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages