Skip to content

Chroma

Runs inference of a Chroma model for protein structure generation, supporting both conditional and unconditional modes with extensive control over SDE integration parameters.

Chroma

Quick Start

  1. Set the desired number of integration steps and chain length.
  2. (Optional) Provide an initial PDB for conditional generation.
  3. Adjust SDE and Langevin parameters as needed.
  4. Run the node to generate protein structures.

Setup Guide

1. Basic Configuration

  1. Choose the number of integration steps (steps) and chain length (chain_length).
  2. Set SDE parameters such as tspan, langevin_factor, and inverse_temperature.
  3. (Optional) Provide an initial PDB for conditional generation.

2. Advanced Options

  1. Select integration and SDE functions (integrate_func, sde_func).
  2. Configure output options like trajectory_length and full_output.
  3. Use chain manipulation options if needed (swap_chain_names, order_chain_names).

Basic Usage

Unconditional Generation

  • Leave input_pdb empty to generate new protein structures from scratch.
  • Set chain_length and steps to control output size and detail.

Conditional Generation

  • Provide an input_pdb to guide the generation process.
  • Use chain manipulation options to adjust output chain order or names.

Configuration

Required Inputs

Field Description Type Example
steps Number of integration steps for the SDE. INT 500
chain_length Length of the protein chain. INT 100
tspan Time span for SDE integration (comma-separated floats). STRING 1.0,0.001
langevin_factor Strength of Langevin noise. FLOAT 2.0
langevin_isothermal Use isothermal Langevin SDE. BOOLEAN False
inverse_temperature Inverse temperature parameter for SDE. FLOAT 10.0
initialize_noise Initialize noise for SDE integration. BOOLEAN True
integrate_func Integration function to use. COMBO euler_maruyama
sde_func SDE function to use. COMBO reverse_sde
trajectory_length Number of steps in trajectory output (max: steps). INT 200
full_output Return full SDE integration outputs. BOOLEAN False
swap_chain_names Comma-separated chain names to swap after generation. STRING A,B
order_chain_names Comma-separated chain names to specify output order. STRING A,B
seed Base seed for randomness. INT 42
mode Execution mode: MOCK, PROD, or TEST. COMBO PROD

Optional Inputs

Field Description Type Example
input_pdb Initial PDB to start generation from. PDB {"1abc": "..."}

Outputs

Field Description Type Example
generation.pdb Generation result. PDB {"gen_0": "..."}
trajectory.pdb Trajectory PDB (if full_output is True). PDB {"traj_0": "..."}
Xhat_trajectory.pdb Xhat trajectory PDB (if full_output is True). PDB {"xhat_0": "..."}
Xunc_trajectory.pdb Xunc trajectory PDB (if full_output is True). PDB {"xunc_0": "..."}

Best Practices

Parameter Selection

  • Start with default parameters for most use cases.
  • Increase steps and trajectory_length for higher resolution outputs.

Reproducibility & Modes

  • Use the same seed for reproducible results.
  • Use MOCK mode for workflow validation, TEST for quick checks, and PROD for full runs.

Troubleshooting

Common Issues

  • Output is empty or missing: Ensure all required parameters are set and input PDB (if used) is valid.
  • Long runtime: Reduce steps or use TEST mode for faster execution.
  • Chain manipulation not working: Check that chain names are valid and comma-separated.

Need Help?

  • Contact support for further assistance.