sample

REINVENT4 Molecular Sampling

Generate molecules by writing a TOML config and running the reinvent CLI directly. No MCP server needed — requires reinvent on PATH (activate the reinvent4 conda env first).

Workflow

1. Resolve paths

readlink -f <relative_path>

All paths in the TOML must be absolute.

2. Determine parameters

• model_file (required): absolute path to .prior or .model file. Use $0 if provided.
• num_smiles: molecules to generate (default 100). Use $1 if provided.
• generator: reinvent (de novo), libinvent (scaffold decoration), linkinvent (fragment linking), mol2mol (analogue generation). Default reinvent.
• smiles_file: required for libinvent / linkinvent / mol2mol
• sample_strategy: multinomial (default, diverse) or beamsearch (deterministic, high-likelihood)
• temperature: sampling temperature (default 1.0); lower → more conservative, higher → more diverse
• unique_molecules: deduplicate output (default true)
• randomize_smiles: randomise SMILES representation for augmentation (default true)
• device: cpu (default) or cuda:0

3. Create the output directory

Choose a timestamped workdir, e.g. <model_parent>/reinvent_runs/sampling_<YYYYMMDD_HHMMSS>/:

mkdir -p <workdir>

4. Write the TOML config

Write <workdir>/config.toml with the Write tool:

run_type = "sampling"
device = "cpu"           # change to "cuda:0" if GPU available

[parameters]
model_file  = "/absolute/path/to/model.prior"
output_file = "/absolute/workdir/sampling.csv"
num_smiles        = 100
unique_molecules  = true
randomize_smiles  = true
# smiles_file      = "/absolute/path/to/input.smi"  # libinvent/linkinvent/mol2mol only
# sample_strategy  = "beamsearch"   # multinomial (default) or beamsearch
# temperature      = 1.0            # lower = conservative, higher = diverse

Show the complete config to the user. Let them request changes before running.

5. Run