schrodinger.application.mlff.inference module

schrodinger.application.mlff.inference.create_jaguar_mlff_model(model_name: str, solvent: Solvent | None = None, solv_model: SolventModel | None = None, execution_device: MlffDeviceType = 0, precision: MlffPrecision = 1, do_force: bool = True, do_virial: bool = False, energy_unit: MlffEnergyUnit = 0, use_base_energies: bool = True) → MlffModel

Create a MLFF model from a given model name. This is a wrapper around the create_mlff function from the energy module. The default settings are chosen to mirror how MLFF models are run through Jaguar.

Parameters:

  • model_name – Model name

  • solvent – Solvent enum. Will fail if the solvent is not Solvent.NONE or None and the model does not support solvent (i.e. isn’t a TB model).

  • solv_model – Solvent model enum value or None. If solvent is set, will default to SolventModel.ALPB. Otherwise, defaults to None.

  • execution_device – Device (CPU or CUDA) on which the model should be executed

  • precision – Floating point precision to use for model inference

  • do_force – Whether the model should compute forces on each atom

  • do_virial – Whether the model should compute the virial tensor. Used in computing the stress tensor.

  • energy_unit – Unit for energy predictions

  • use_base_energies – Toggle inclusion of base atomic energies in the model’s energy predictions. Only some models (e.g. MPNICE) support toggling the inclusion of atomic energies. If unsupported, this keyword will have no effect.

Returns:

MLFF model

Raises:

ValueError – If solvent is not Solvent.NONE/None and the model does not support solvent.

schrodinger.application.mlff.inference.compute_mlff_single_points_structures(sts: list[schrodinger.structure._structure.Structure], predictor: energy.QRNNPredictor | energy.MlffModel, return_failed: bool = True, logger=<built-in function print>) → list[Structure]

Compute single-point energies for a list of structures using a given QRNNPredictor or MlffModel. The input structures are modified in place with output from the model evaluation

The order of precedence for determining the charge of each structure is: 1. Structure charge from PROPERTY_KEY_CHARGE 2. Total charge of the structure determined from formal charges

The order of precedence for determining the multiplicity of each structure is: 1. Structure multiplicity from PROPERTY_KEY_MULT 2. Parity of the number of electrons, taking into account the total charge

Parameters:

  • sts – list of structures

  • predictor – QRNNPredictor or MlffModel instance

  • return_failed – If True, return all structures, even those that failed to compute the MLFF energy.

  • logger – Callable to process any RuntimeErrors encountered

Returns:

list of output structures, with properties added from model output