Skip to main content

    RXNFP quickstart

    Fingerprint one reaction SMILES or a batch from a retrosynthetic route, then search the reference index for the most similar known reactions.

    You want to turn a reaction into a fingerprint vector, or find known reactions similar to one you have. This runs both paths in a few steps: submit reaction SMILES, review the fingerprint, or run a similarity search against the reference index.

    RXNFP quickstart

    This guide covers fingerprinting a single reaction, batch-fingerprinting reactions from a retrosynthetic route, and searching for similar known reactions.

    Workflow 1: Fingerprint a single reaction

    Submit the reaction

    Navigate to Tools > RXNFP in .bloom*, or ask Ivy:

    Generate an RXNFP fingerprint for: CC(=O)Cl.OCC>>CC(=O)OCC

    Review the fingerprint

    The result includes the 256-dimensional vector, plus is_valid and any per-reaction error. In the canvas card, a compact visualization shows the fingerprint as a heatmap strip.

    Workflow 2: Batch fingerprint retrosynthetic steps

    Submit the batch

    Provide reaction SMILES from an AiZynthFinder route as an array:

    {
      "reaction_smiles": [
        "c1ccc(B(O)O)cc1.c1ccc(Br)cc1>>c1ccc(-c2ccccc2)cc1",
        "CC(=O)Cl.c1ccccc1N>>CC(=O)Nc1ccccc1",
        "CC(=O)O.OCC>>CC(=O)OCC"
      ]
    }

    Export the fingerprints

    Download the results (CSV or JSON) and use the fingerprint vectors for your own clustering or ML pipeline — the platform returns the vectors, not a built-in clustering view.

    Workflow 3: Find similar known reactions

    Find the 5 most similar known reactions to: CC(=O)Cl.OCC>>CC(=O)OCC

    Or configure directly:

    ParameterValue
    reaction_smilesThe query reaction
    top_k5
    min_similarity0.7 (default)

    Review the matches

    Each match reports its cosine similarity to the query, its SMILES (when available), and its reaction name/classification (when available) from the reference index — see outputs & interpretation.

    Next steps

    Was this page helpful?

    On this page

    On this page

    Loading...