heracleum-tox-mcp-server

An MCP server that reproduces the results of:

Rassabina, A.E.; Fedorov, M.V. Analysis of the Toxicological Profile of Heracleum sosnowskyi Manden. Metabolites Using In Silico Methods. Plants 2025, 14, 3253. https://doi.org/10.3390/plants14213253

The paper runs entirely on the proprietary Syntelly platform. Since Syntelly is not openly accessible, this server reproduces the same pipeline with open-source analogues of every Syntelly module — including the very models Syntelly itself uses (fingerprint-based CatBoost + fragment-based XGBoost; Sosnin et al., Molecules 2024, 29, 1826, the platform paper cited as ref. [36]), trained on the same open datasets the paper names (TOXRIC / ChemIDplus / PyTDC).

Syntelly → open-source analogue mapping

Related MCP server: Comptox MCP Server

Tools

Each tool returns {"answer": ..., "metadata": ...}. Figures are saved as PNG to a local artifacts dir (HERACLEUM_ARTIFACTS_DIR) or, if S3 is configured, uploaded and returned as presigned URLs (same pattern as chemical-mcp-server / tox-antitargets-mcp-server).

Reproduction fidelity

reproduce_all and pytest tests/ assert these against the paper:

The full 225-compound dataset (standardized SMILES + SynID + cluster A–E) is reproduced exactly from the paper's Supplementary Tables S1–S5 — parsed by parse_supplementary.py into server/data/supplementary_smiles.csv, then assembled by build_dataset.py (which merges the cluster-E Table 2 toxicity values and resolves compound names via PubChem). The paper's own model-quality numbers (Supplementary Table S6) are bundled for comparison (model_quality / reproduce_all).

Documented open-analogue divergences (faithful method; the small open datasets disagree with Syntelly's proprietary models):

• DILI / hepatotoxicity: the open TDC DILI model (n=475) predicts most cluster-E furanocoumarins as non-hepatotoxic, opposite to Syntelly's "all DILI-toxic". The applicability domain flags these as moderate-reliability — an honest signal that the open set under-covers furanocoumarins. This is the one paper claim (C5) that does not reproduce, and it is reported as such.

• Cardiotoxicity: the open hERG proxy is more conservative than Syntelly's cardiotox model (it flags furanocoumarins as hERG blockers; the paper found none).

• Per-route LD50: TOXRIC's six per-route mouse sets are not openly scriptable, so all routes share the open acute-LD50 model unless you supply per-route CSVs (see below). The cluster ranking (E most toxic) is the robust open reproduction.

The clustering uses a differential fingerprint (ECFP of the Bemis–Murcko scaffold) as the open analogue of SynMap's differential fingerprints + parametric t-SNE (Karlov/Sosnin/Tetko/ Fedorov, ACS Omega 2021). Emphasising the core scaffold separates furanocoumarins (E) from simple aromatics (D), recovering all five families at ~95 % agreement; set HERACLEUM_CLUSTER_FINGERPRINT=ecfp4 for the plain-molecule fallback (~86 %, merges D into E).

Run locally

git clone https://github.com/chemagents/heracleum-tox-mcp-server
cd heracleum-tox-mcp-server
cp .env.example .env
uv sync
uv pip install --no-deps "PyTDC==0.4.1"     # open datasets; pins old rdkit-pypi, so --no-deps
uv run python prepare_models.py             # train & cache the open models (downloads TDC data)
uv run python -m server.heracleum_server    # serves http://0.0.0.0:7331/mcp

# The 225-compound dataset is already bundled (server/data/heracleum_metabolites.csv).
# To regenerate it from the paper's Supplementary PDF:
#   pdftotext -layout plants-3875800-supplementary.pdf supp.txt
#   uv run python parse_supplementary.py supp.txt   # -> server/data/supplementary_smiles.csv
#   uv run python build_dataset.py                  # merges Table 2 refs + PubChem names

Run with Docker

docker compose up -d --build      # host port 7336 -> container 7331

To run it inside the CoScientist stack instead, add this repo as a service in mcp-servers/docker-compose.yml (the CoScientist repo already includes such an entry).

The Docker build installs PyTDC and pre-trains the models (best-effort; if there is no network at build time the server trains them lazily on first request).

Attach to CoScientist

CoScientist discovers MCP tools via RAG (Postgres + Qdrant). Register this server once:

# from the CoScientist repo root, with the RAG stack running and .env configured
python scripts/rag_tools/cli.py load mcp-servers/heracleum-tox-mcp-server/rag_registration.json
# or directly:
python scripts/rag_tools/cli.py add \
  --url http://localhost:7336/mcp \
  --name heracleum-tox \
  --description "In-silico toxicology of Heracleum sosnowskyi metabolites; LD50, hepato/DILI/cardio/carcinogenicity, furanocoumarins (Rassabina & Fedorov 2025)"

After registration the ToolRetrieverAgent surfaces these tools for plant-metabolite / toxicity / LD50 / furanocoumarin queries, and ExperimentAgent (FEDOT.MAS) calls them by URL. If CoScientist runs in the same Docker network, register the in-network URL instead: http://heracleum-tox-mcp-server:7331/mcp.

See REPRODUCTION_QUESTIONS.md for the exact prompts to ask CoScientist (one per paper assertion, plus a single "reproduce everything" prompt).

Exact per-route LD50 reproduction (optional)

The paper predicts LD50 for six mouse routes from TOXRIC. To reproduce those exactly, drop TOXRIC per-route CSVs (smiles,y with y = -log10(mol/kg)) named ld50_<route>.csv (oral,iv,ip,sc,skin,im) into HERACLEUM_LD50_DATA_DIR; route-specific models then train automatically.

Tests

uv run pytest tests -v                 # all (trains models on first run, then cached)
uv run pytest tests -v -m "not slow"   # fast deterministic checks only

License / data

Open datasets via Therapeutics Data Commons (PyTDC) and TOXRIC. Please cite Rassabina & Fedorov (2025) when using these results, and TDC / the Syntelly platform paper (Sosnin et al., Molecules 2024, 29, 1826) for the methods.