Proteinmpnn Tools#

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

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

Available Tools#

proteinmpnn_design (Type: RemoteTool)#

Design protein SEQUENCES for a fixed backbone with ProteinMPNN (inverse folding). Give the backbo…

proteinmpnn_design tool specification

Tool Information:

  • Name: proteinmpnn_design

  • Type: RemoteTool

  • Description: Design protein SEQUENCES for a fixed backbone with ProteinMPNN (inverse folding). Give the backbone as PDB text; returns num_sequences designed sequences with each one’s score and its sequence recovery against the structure’s own sequence. Raise ‘temperature’ for diversity. ‘weight_set’ selects ProteinMPNN (vanilla) or SolubleMPNN. Reproduces upstream’s released protein_mpnn_run.py field for field at the same seed, which is checked by byte-comparison in this provider’s acceptance suite. Runs on the CPU and holds no GPU memory, so it can share a host with any folding model: expect ~20 s for 8 sequences of a 106-residue backbone on a quiet 20-core box, roughly linear in both length and sequence count.

Parameters:

  • structure (string) (required) The backbone, as the text of a PDB file. ATOM lines of amino-acid residues are read (N, CA, C, O); waters, ligands and side chains are ignored. Every chain in the file is designed. Up to 8 MB and 3000 residues.

  • num_sequences (integer) (optional) How many sequences to design (default 8, max 128). Must be a whole number of batches of ‘batch_size’.

  • batch_size (integer) (optional) How many sequences are decoded per forward pass (default: all of them). Upstream’s own option; lower it only to hold memory down on a long backbone.

  • temperature (number) (optional) Sampling temperature, in (0, 1]. Upstream’s default is 0.1, which is close to greedy; 0.2-0.3 trades score for diversity.

  • weight_set (string) (optional) ‘vanilla’ is ProteinMPNN as released and upstream’s default. ‘soluble’ is SolubleMPNN, the same architecture trained without membrane proteins; use it for soluble designs.

  • seed (integer) (optional) Random seed, >= 1 (default 37). The same seed and the same backbone give the same sequences; the worker will not draw one for you.

Example Usage:

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