Af3 Torch Tools#

Configuration File: remote_tools/af3_torch_tools.json Tool Type: Remote Tools Count: 1

This page contains all tools defined in the af3_torch_tools.json configuration file.

Available Tools#

af3_torch_predict (Type: RemoteTool)#

Predict the all-atom structure of a biomolecular complex with AF3-torch (the xfold PyTorch port o…

af3_torch_predict tool specification

Tool Information:

  • Name: af3_torch_predict

  • Type: RemoteTool

  • Description: Predict the all-atom structure of a biomolecular complex with AF3-torch (the xfold PyTorch port of AlphaFold 3, running the public OpenFold3-preview2 parameters). Input is an AlphaFold 3 fold-input JSON (dialect ‘alphafold3’), and every chain must carry its own MSAs inline: no genetic-database search is run. Returns the paths of the predicted mmCIF files and their confidence and ranking scores; the mmCIF text itself only when return_structures is true.

Parameters:

  • fold_input (string) (required) An AlphaFold 3 fold-input JSON, as text. Example:

{“name”: “x”, “dialect”: “alphafold3”, “version”: 2, “sequences”: [{“protein”: {“id”: “A”, “sequence”: “MKTAYIAKQRQ…”, “modifications”: [], “templates”: [], “unpairedMsa”: “>querynMKTAYIAKQRQ…n”, “pairedMsa”: “”}}], “modelSeeds”: [1]} modelSeeds chooses the sampler seeds; one prediction runs per seed. unpairedMsa may be the query alone for a single-sequence prediction, but accuracy is much lower than with a real MSA.

  • name (string) (optional) A job name used for the output file names (default ‘fold’). Letters, digits, ‘-’ and ‘_’.

  • return_structures (boolean) (optional) Return the mmCIF text of the best-ranked model (default false: it is megabytes for a real complex, and the transport truncates a reply at 100,000 characters). Paths and scores come back either way.

Example Usage:

query = {
    "name": "af3_torch_predict",
    "arguments": {
        "fold_input": "example_value"
    }
}
result = tu.run(query)