The Force-Field Parameterizer prepares small molecules for molecular simulation. It assigns an OpenFF Sage force field and partial charges, checks that every bond, angle, torsion, and van der Waals term was assigned, and writes each molecule as a system file with a record of exactly which force field and charge model were used. Simulations downstream then reuse one checked parameter set instead of regenerating it each time.
It assigns parameters only: no dynamics, no quantum chemistry. It runs on CPU.
How it works
Using the OpenFF Toolkit and Interchange (with RDKit and AmberTools), for each molecule it:
- reads the 3D structure, which needs explicit hydrogens;
- checks formal charge, elements, radicals, and stereochemistry;
- assigns partial charges (
charge_method); - assigns the force field (
force_field) and confirms nothing was left unassigned; - writes the system for each simulation engine you ask for.
Force fields: OpenFF releases openff-2.0.0 to openff-2.3.0; the default, openff-2.2.0, matches what OpenMM MD and MM-GBSA Rescoring use.
Charges: am1bcc (the default, the model Sage was fitted with), nagl (a fast machine-learned model trained to reproduce AM1-BCC), gasteiger, or user (charges already in your SDF or MOL2). Results using gasteiger or user charges carry a warning, because Sage wasn't fitted with them.
Inputs
Up to 100 molecules as SDF, MOL2, or MOL blocks with explicit hydrogens and 3D coordinates. Supported elements are H, C, N, O, F, P, S, Cl, Br, and I, with up to 100 heavy atoms per molecule. Undefined stereocenters are rejected unless you set allow_undefined_stereo.
Outputs
- Per molecule: an OpenMM
system.xml(always), the topology, the serialized Interchange, and GROMACS (.top,.gro) or AMBER (.prmtop,.inpcrd) files if you added those toengines. - Charged molecules as SDF (and MOL2 where possible).
coverage.csv: one row per molecule with its status and, if it failed, why (for examplemissing_hydrogens,unsupported_element,charge_sum_mismatch,unassigned_parameters).charges.csv: every atom's formal and partial charge.provenance.json: the force field file and its SHA-256, the charge model, and the toolkit versions.
A molecule that fails a check is reported and skipped; the job fails only if every molecule fails.
Related tools
Use the parameterized systems in OpenMM MD. To make 3D structures from SMILES first, use Molecule Conversion or Conformer Ensemble Generator.