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 of two packages:

  • graph/, the package bpgraph: 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.
  • client/, the package bpgraph-client: what a repository consulting the graph needs — its agent's instructions, the analysis skills, and the command that does their mechanical part. It 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                     # both packages

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 client/README.md how to consult 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