Advanced Crystal Structure Prediction Workflows

Please note that Job Server is required for submission of crystal structure prediction jobs. To learn more about Job Server, see Job Submission to a Cluster.

 

This section highlights how advanced crystal structure prediction (CSP) workflows may be set up and run. For an overview of running CSP from Maestro and the command line, see the Crystal Structure Prediction Panel and Crystal Structure Prediction Workflow respectively.

Advanced Crystal Structure Prediction Workflows Table of Contents

Z′=2

Crystal structure predictions for Z′=2, two independent molecules in an asymmetric unit (ASU), is a significantly more complex workflow involving larger scale calculations than for Z′=1, a single molecule in an asymmetric unit. A pair of conformers can adopt many different relative geometries to serve as the ASU of the Z′=2 crystal structures.

To make the calculations scalable, we limit the conformer pool for Z′=2 CSP to a subset of the lowest energy conformers from Z′=1 CSP. This is based on the assumption that conformers in the stable Z′=2 crystal structures should have low strain energies, and these low energy conformers have corresponding low energy Z′=1 structures. We have validated this assumption on a large number of molecules.

To run the CSP workflow for Z′=2:

  1. Conformer Generation (ConfGen): Conformers are generated for the Z′=2 workflow based on the output of the corresponding Z′=1 CSP calculation. To run a Z′=1 CSP calculation, see the Crystal Structure Prediction Panel or Crystal Structure Prediction Workflow. The Z′=2 search is restricted to a subset of the lowest energy conformers identified during the Z′=1 DFT relaxation. To extract these conformers, run the following command:

    $SCHRODINGER/run -FROM scisol prepare_conformer_pool_from_zp1_output.py output_dft_relaxation.maegz Conformers_For_ZP2.mae

    Where output_dft_relaxation.maegz is the output file from the last DFT stage of Z′=1 prediction.

    The subsequent Packing Search stage accepts a maximum of 20 conformers. If more than 20 unique conformers are produced, use the -energy-window option to narrow the pool, or trim it before running Packing Search.

  2. Packing Search (PackingSearch): Create crystal packings of the various Z′=2 conformers. Generate the default configuration file as you would for Z′=1 calculations and make the specified changes to the file:

    $SCHRODINGER/run -FROM scisol csp/packing_search.py -write-default-config config.yaml

    Edit the config.yaml file as follows:

    Keyword Setting Description
    conformer_pool Conformers_For_ZP2.mae The structure file containing the Z′=2 conformers created in the previous step.
    is_zprime2_search true Enables the Z′=2 specific search.
    zprime2_search_modes dimer-first, rod, or layer Z′=2 has three complementary search modes. Running all three simultaneously is possible but is not recommended due to the high memory and scratch space required. It is suggested to run each mode as a separate job, and then merge the output structures of the packing search from each mode together as the input for the next MD stage.

    Alternatively, run the combined Workflow (csp_packing_to_ranking.py) to perform Packing Search and the three energy-ranking stages (MD, MLFF, and DFT Relaxation) in a single job.

  3. Molecular Dynamics (MD) Relaxation (MDRelaxation): A molecular dynamics simulation is run on each crystal structure generated by the packing search. The outputs with high energy, >10 kcal/mol relative to the lowest energy structure, are removed prior to the next stage of ranking. Generate the configuration file as you would for Z′=1 calculations and run the calculation.

  4. Machine Learning Force Field (MLFF) Relaxation (MLFFRelaxation): The crystal structures are then relaxed using MLFF and the outputs with high energy, >6 kcal/mol relative to the lowest energy structure, are removed. Generate the configuration file as you would for Z′=1 calculations and run the calculation.

  5. Density Functional Theory (DFT) Relaxation (DFTRelaxation): Relaxes the crystals using periodic DFT (first with PBE-D3 functional , then with R2SCAN-D3 functional) and removes those with high energy. Generate the configuration file as you would for Z′=1 calculations and run the calculation.

Monohydrates/Solvates/Co-crystals

Crystal structure predictions for monohydrates, solvates, and co-crystals involve an organic molecule paired with an excipient (water, a solvent, or a coformer) in the asymmetric unit (ASU). These are more complex workflows involving larger scale calculations than for Z′=1, a single organic molecule in the ASU. The workflow steps are the same for all three types; the only difference is the excipient specified in the packing search configuration.

  1. Conformer Generation (ConfGen): Conformers are generated for the monohydrate/solvate/co-crystal CSP workflow based on the output of the corresponding Z′=1 CSP calculation. To run a Z′=1 CSP calculation, see the Crystal Structure Prediction Panel or Crystal Structure Prediction Workflow. The search is restricted to a subset of the lowest energy conformers identified during the Z′=1 DFT relaxation. To extract these conformers, run the following command:

    $SCHRODINGER/run -FROM scisol prepare_conformer_pool_from_zp1_output.py output_dft_relaxation.maegz Conformers_For_Excipient.mae

    Where output_dft_relaxation.maegz is the output file from the last DFT stage of Z′=1 prediction.

    The subsequent Packing Search stage accepts a maximum of 20 conformers. If more than 20 unique conformers are produced, use the -energy-window option to narrow the pool, or trim it before running Packing Search.

  2. Packing Search (PackingSearch): Create crystal packings of the conformers with the chosen excipient. Generate the default configuration file as you would for Z′=1 calculations and make the specified changes to the file:

    $SCHRODINGER/run -FROM scisol csp/packing_search.py -write-default-config config.yaml

    Edit the config.yaml file as follows:

    Keyword Setting Description
    conformer_pool Conformers_For_Excipient.mae The structure file containing the conformers created in the previous step.
    add_supported_excipients e.g. water, methanol, Succinicacid Specify the excipient to pair with the organic molecule in the ASU. Use water for monohydrates, a solvent name for solvates, or a coformer name for co-crystals. Run $SCHRODINGER/run print_csp_solv_conf.py for the full list of supported solvents and coformers.

    Alternatively, run the combined Workflow (csp_packing_to_ranking.py) to perform Packing Search and the three energy-ranking stages (MD, MLFF, and DFT Relaxation) in a single job.

  3. Molecular Dynamics (MD) Relaxation (MDRelaxation): A molecular dynamics simulation is run on each crystal structure generated by the packing search. The outputs with high energy, >10 kcal/mol relative to the lowest energy structure, are removed prior to the next stage of ranking. Generate the configuration file as you would for Z′=1 calculations and run the calculation.

  4. Machine Learning Force Field (MLFF) Relaxation (MLFFRelaxation): The crystal structures are then relaxed using MLFF and the outputs with high energy, >6 kcal/mol relative to the lowest energy structure, are removed. The default MLFF model (Organic_Crystals_MPNICE_v1.1) has been trained to better describe interactions in multicomponent crystals, as well as increase the coverage of element space for common monatomic counterions including halogens (F, Cl, Br, I), Alkali metals (Na, K), and Alkali earth metals (Mg, Ca).

    Generate the configuration file as you would for Z′=1 calculations and run the calculation.

  5. Density Functional Theory (DFT) Relaxation (DFTRelaxation): Relaxes the crystals using periodic DFT (first with PBE-D3 functional, then with R2SCAN-D3 functional) and removes those with high energy. Generate the configuration file as you would for Z′=1 calculations and run the calculation.

Salts

Crystal structure predictions for salts, a charged molecule in an asymmetric unit (ASU), is a complex workflow requiring some changes when compared to the workflow for Z′=1, an uncharged single molecule, in an asymmetric unit (ASU).

  1. Conformer Generation (ConfGen): Conformers are generated for the salt CSP workflow in the same way as Z′=1 CSP calculations. See the Crystal Structure Prediction Panel or Crystal Structure Prediction Workflow.to generate the conformers.

  2. Packing Search (PackingSearch): Create crystal packings of the various salt conformers. Generate the default configuration file as you would for Z′=1 calculations and make the specified changes to the file:

    $SCHRODINGER/run -FROM scisol csp/packing_search.py -write-default-config config.yaml

    Edit the config.yaml file as follows:

    Keyword Setting Description
    conformer_pool Conformers_For_Salts.mae The renamed structure file containing the salt conformers created in the previous step.
    add_supported_excipients Br-, Cl-, F-, I-, K+, or Na+ Specify the counterion to balance the charged molecule in the ASU.

    Alternatively, run the combined Workflow (csp_packing_to_ranking.py) to perform Packing Search and the three energy-ranking stages (MD, MLFF, and DFT Relaxation) in a single job.

  3. Molecular Dynamics (MD) Relaxation (MDRelaxation): A molecular dynamics simulation is run on each crystal structure generated by the packing search. The outputs with high energy, >10 kcal/mol relative to the lowest energy structure, are removed prior to the next stage of ranking. Generate the configuration file as you would for Z′=1 calculations and run the calculation.

  4. Machine Learning Force Field (MLFF) Relaxation (MLFFRelaxation): The crystal structures are then relaxed using MLFF and the outputs with high energy, >6 kcal/mol relative to the lowest energy structure, are removed. Salt crystal structures have strong electrostatic interactions, which is a different chemical space than organic neutral molecules. The default MLFF model (Organic_Crystals_MPNICE_v1.1) has been trained to better describe these electrostatic interactions, as well as increase the coverage of element space for common monatomic counterions including halogens (F, Cl, Br, I), Alkali metals (Na, K), and Alkali earth metals (Mg, Ca). This model is expected to have enhanced performance for energetically ranking hydrate crystals.

    Generate the configuration file as you would for Z′=1 calculations and run the calculation.

  5. Density Functional Theory (DFT) Relaxation (DFTRelaxation): Relaxes the crystals using periodic DFT (first with PBE-D3 functional , then with R2SCAN-D3 functional) and removes those with high energy. Generate the configuration file as you would for Z′=1 calculations and run the calculation.