deepretro.metadata

LLM-assisted helpers for reaction metadata prediction.

The module exposes the ported metadata agents for reagent, condition, and literature-reaction prediction. Functions return (status_code, payload) tuples for compatibility with the legacy metadata workflow and use package imports, structured logging, and in-memory cache helpers.

recommend_reaction_metadata is the high-level entry point for reaction SMILES strings. It parses reactants>agents>products or reactants>>products input, recommends reagents, predicts reaction conditions, and retrieves the nearest literature reaction.

Example

from deepretro.metadata import reagent_agent

status, reagents = reagent_agent(
    reactants=[{"smiles": "CCO"}],
    product=[{"smiles": "CC=O"}],
    model="openai/gpt-4o-mini",
)

from deepretro.metadata import recommend_reaction_metadata

status, metadata = recommend_reaction_metadata(
    "CCO.CC(=O)O>>CCOC(C)=O",
    model="openai/gpt-4o-mini",
)

API Reference