Full biomaterials graph — 1.57M nodes (prebuilt TuringDB LOAD JSONL)

Prebuilt full graph in TuringDB APOC LOAD JSONL format — skip the 810MB source download and the rebuild.

1,573,076 nodes (1,495,668 Compound · 72,291 Reaction · 5,105 Enzyme · 6 Polymer · 6 Property) and 363,624 edges (SUBSTRATE_OF/PRODUCES carbon backbone, USES_COFACTOR, CATALYZES, POLYMERIZES_TO, HAS_PROPERTY).

Which file?

• graph.jsonl (slim, 512MB) — full properties on the ~120k-node connected metabolic core; the ~1.45M isolated catalogue compounds keep identity fields (id, name, InChIKey) but not SMILES/formula/mass. Loads in ~15s and fits on an 8GB machine. Recommended for most people.
• graph_full_props.jsonl (rich, 829MB) — SMILES/formula/charge/mass on every compound. Needs ~16GB+ RAM to load (it OOMs an 8GB box).

Load

turingdb start -demon
cp graph.jsonl ~/.turing/data/          # or graph_full_props.jsonl
# then, via the Python SDK:
#   client.query("LOAD JSONL 'graph.jsonl' AS biomaterials")
#   client.set_graph("biomaterials")

Or just python load/load_full_jsonl.py from the repo.

Both regenerate from source with python data/expand_from_retrorules.py --full [--rich]. Data: RetroRules + MetaNetX/MNXref (CC-BY, see NOTICE).