mlff.py Command Help
Command: $SCHRODINGER/jaguar run mlff.py
usage: $SCHRODINGER/jaguar run mlff.py [-h] [-WAIT] [-DEBUG] [-jobname <name>]
[-subdir] [-recover] [-no_subjob_files]
[-scr <absolute path>] [-PARALLEL <N>]
[-max_threads <T>]
[-procs_per_node <N>]
[-gpu_host <host>:<M>]
[-use_one_node | -use_multiple_nodes]
[-HOST <host>:<M>]
[-SUBHOST <host>:<M>] [-SAVE]
[-NOJOBID] [-OPLSDIR <oplsdir>]
[-MLFF_DATA <mlff_data_dir>]
infile
positional arguments:
infile Single MLFF .in file
options:
-h, --help show this help message and exit
other options:
-WAIT Wait for job to finish before returning prompt.
-DEBUG, -D Print detailed information about job launch.
-jobname <name> Set the job name.
-subdir Run Jaguar in a sub-directory.
-recover Manually re-run a job using the recover mechanism. NOTE this option is not recommended for default recovery jobs.
Use "jaguar run <filename>.recover" instead (see documentation for more details).
-no_subjob_files Do not return subjob output files to launch directory.
-scr <absolute path> Specify a scratch directory (must not already exist). Directory must be given as an absolute path.
Note this will be used by the Fortran backend and is independent of the specification of -TMPDIR.
-PARALLEL <N> Use up to <N> CPUs simultaneously for the whole workflow, automatically allocated among subjobs, including threaded subjobs.
-max_threads <T> Use no more than <T> OpenMP threads for each Jaguar subjob. Default 8.
-procs_per_node <N> Use no more than <N> CPUs per node. Default is taken from the schrodinger.hosts file; if undefined 8.
-gpu_host <host>:<M> Host on which to run Jaguar subjobs that request GPU. The optional :<M> defines the maximum number of simultaneous GPU subjobs. Defaults to the value resolved from -SUBHOST/-HOST. Validated only when a GPU subjob is actually launched.
-use_one_node Force CPU resources to be requested upfront on one node.
This pool of CPUs will be used for the duration of the job instead of resubmitting to the queue.
-use_multiple_nodes Force CPU resources to be requested dynamically from the queue (if available) instead of upfront on one node.
commonly used Schrodinger Suite options:
-HOST <host>:<M> Run job remotely on host <hostname>. The optional :<M> defines the maximum number of simultaneous subjobs.
May be combined with -PARALLEL <N>.
-SUBHOST <host>:<M> Run any subjobs remotely on subhost <hostname>. The optional :<M> defines the maximum number of simultaneous subjobs.
May be combined with -PARALLEL <N>.
-SAVE Return .zip file of scratch directory.
-NOJOBID Run Jaguar interactively without jobserver (not available with python workflows).
-OPLSDIR <oplsdir> Use custom FF parameters from specified directory for workflows which support it.
-MLFF_DATA <mlff_data_dir>
Enable custom MLFF models from specified directory for workflows which support it.
mlff Input Keywords:
----------------------------------------
infile: Input .mae file containing one or more structures.
Default=
solvent: Solvent to use for implicit solvation. Valid solvents include:
water, acetonitrile, aniline, benzene, dcm, chloroform, cs2, dmf, dmso, diethyl ether, ethanol, ethyl acetate, hexadecane, hexane, nitromethane, 1-octanol, toluene, thf.
Default=water
solvent_model: Solvent model to use for implicit solvation. Note: `cpcmx` is not supported for optimizations.
Default=alpb
model: Model to use for the MLFF calculation. Commonly used models include:
Organic_MPNICE: Message Passing model for organic molecules.
Organic_MPNICE_TB: Delta-Learned Message Passing model for organic molecules.
UMA_sm_omol: Universal Model for Atoms, trained on OMol25.
For the full list of available models, see the Machine Learning Force Field documentation.
Default=qrnn_3
charge: Total charge of each molecule. If not specified, the charge will be calculated.
Default=None
multiplicity: Multiplicity of each molecule. If not specified, the multiplicity will be calculated.
Only the UMA_* models support `multiplicity>1`.
Default=None
batches: Split the input molecules into this many batchesto be submitted as separate subjobs.
If not set, the number of batches will be automatically determined
to best utilize requested CPUs resources.
Default=None
opt_steps: Perform at most this many geometry optimizations steps
Default=1000
gpu: Whether to use GPU acceleration for the MLFF calculation. Only available for some models (QRNN is not supported).
Delta-learned models will not benefit significantly from GPU acceleration
as the xTB portions of the calculation are still performed on the CPU.
Default=False
implicit_solvent: Add implicit solvation.
Can only be used with the Delta-Learned models: (<QRNNModels.DELTA: 2>, <QRNNModels.DELTA_TRANSFER: 3>, 'qrnn_2', 'qrnn_3').
Default=False
use_base_energies: Determines whether base atomic energies should be included in MLFF predictions. Note that for some models (e.g. QRNN, uma_sm_omol), base energies are always included in the model (un-toggleable).
Default=True
geopt: Perform a geometry optimization on the input structures.
Default=False
preopt: Perform a pre-optimization with OPLS before the MLFF calculation.
Default is False for single-point calculations and True for geometry optimizations.
Default=None
flexible_convergence: Use flexible convergence criteria for the geometry optimization.
Convergence is determined via a point system:
A point is awarded for each of the convergence criteria below the threshold.
For gradient/displacement criteria, an extra point is awarded for reaching 1/5 the threshold.
For the energy difference, an extra point is awarded for reaching 1/20 and 1/200 the threshold.
The optimization is considered converged when the total points reaches 5 (3 with `ignore_displacement`).
Default=True
ignore_displacement: Ignore the displacement convergence criteria in the geometry optimization.
Default=True
return_failed: Return structures for which the geometry optimization or xTB iterations failed.
Default=True
verbose: Print additional information to the output file (e.g. convergence progress for each structure).
Default=False
prepare_input_structures: Preprocess input structures, converting coordinates from 2D to 3D and adding any missing hydrogens.
Default=False