Genie3 Tools#

Configuration File: remote_tools/genie3_tools.json Tool Type: Remote Tools Count: 1

This page contains all tools defined in the genie3_tools.json configuration file.

Available Tools#

genie3_design (Type: RemoteTool)#

Design protein backbones with Genie 3. Either give a ‘length’ to sample unconditional backbones o…

genie3_design tool specification

Tool Information:

  • Name: genie3_design

  • Type: RemoteTool

  • Description: Design protein backbones with Genie 3. Either give a ‘length’ to sample unconditional backbones of that size, or a ‘target’ from the shipped binder-design benchmark to design binders against it. Returns a per-design record; the PDB text itself is opt-in.

Parameters:

  • length (string) (optional) Backbone length for unconditional design: ‘100’, or a range ‘100-200’ sampled in steps. 10-670.

  • length_step (integer) (optional) Step between lengths when ‘length’ is a range (default: the whole range in one step, i.e. one length). n_designs backbones are sampled at EACH length.

  • target (string) (optional) Binder-design target from the pinned checkout’s benchmark: 01_bhrf1, 02_sc2rbd, 03_il7ra, 04_pdl1, 05_trka, 06_insulinr, 07_h1, 08_vegfa, 09_il17a, 10_tnfa.

  • n_designs (integer) (optional) Backbones per length or per target (default 1, max 8).

  • n_steps (integer) (optional) DDIM sampling steps (default 100, upstream’s own; max 200). Fewer is faster and lower quality.

  • batch_size (integer) (optional) Designs per denoiser call (default 1, max 2). Higher is faster per design and costs proportionally more VRAM; the kit refuses a batch that does not fit by name, before it runs.

  • seed (integer) (optional) Random seed; the same seed reproduces the designs.

  • return_structures (boolean) (optional) Return the PDB text of every backbone (default false).

Example Usage:

query = {
    "name": "genie3_design",
    "arguments": {
    }
}
result = tu.run(query)