Af3 Jax Tools#

Configuration File: remote_tools/af3_jax_tools.json Tool Type: Remote Tools Count: 1

This page contains all tools defined in the af3_jax_tools.json configuration file.

Available Tools#

af3_jax_predict (Type: RemoteTool)#

Predict an all-atom biomolecular structure with AlphaFold 3 (the open JAX fork) on public OpenFol…

af3_jax_predict tool specification

Tool Information:

  • Name: af3_jax_predict

  • Type: RemoteTool

  • Description: Predict an all-atom biomolecular structure with AlphaFold 3 (the open JAX fork) on public OpenFold3-preview2 weights. Give either ‘sequences’ – a simple list of chains, which is run MSA-free – or ‘fold_input’, a complete AlphaFold 3 fold-input JSON carrying your own MSAs and templates. No alignment is built here and no external service is contacted. Returns per-sample confidence scores ranked best-first; the mmCIF text itself is opt-in.

Parameters:

  • sequences (array) (optional) The chains of one complex, run MSA-free. Mutually exclusive with ‘fold_input’.

  • fold_input (object) (optional) A complete AlphaFold 3 fold-input JSON object (dialect ‘alphafold3’). Use this when you have real MSAs or templates: every protein chain needs ‘unpairedMsa’, ‘pairedMsa’ and ‘templates’, where an empty string means MSA-free and null means ‘the data pipeline will fill this’, which this provider does not run.

  • seed (integer) (optional) Random seed; replaces ‘modelSeeds’ in a supplied fold_input. The same seed reproduces the prediction.

  • num_diffusion_samples (integer) (optional) Structures to sample per seed (upstream’s default is 5). Each is scored separately, so this is how to compare conformations. Changing it recompiles the held model.

  • num_recycles (integer) (optional) Trunk recycling iterations (upstream’s default is 10). Fewer is faster and less accurate. Changing it recompiles the held model.

  • return_structures (boolean) (optional) Return the mmCIF text (default false: megabytes per sample). The confidence scores come back either way.

Example Usage:

query = {
    "name": "af3_jax_predict",
    "arguments": {
    }
}
result = tu.run(query)