Fast and automated identification of reactions with low barriers using meta-MD simulations

Meta-MD product search procedure

Usage

Procedure to run meta-dynamics search for potential one-step reaction products:

• create .csv file containing the list of mapped reactant SMILES to be investigated
• change parameters CPUS, MEM, MAX_QUEUE, N_RUNS, S_FACTOR, TIME_PS, K_PUSH, ALP, INITIALIZE_PRODUCTS and SCRIPT in control_metadyn_runs.py
  • CPUS: number of cpus each meta-dynamics job can use
  • MEM: amount of memory each meta-dynamics job can use
  • MAX_QUEUE: maximally allowed number of jobs to be submitted to Slurm at the same time
  • N_RUNS: number of meta-dynamics runs to be done for each reactant SMILES
  • S_FACTOR: the scaling parameter for the wall potential
  • TIME_PS: How often to check for change in adjacency matrix in ps
  • K_PUSH: the k_push meta-dynamics parameter
  • ALP: parameter controlling the width of the Gaussian meta-dynamics potential
  • INITILIZE_PRODUCTS: Should the biasing potential be initialized with products from previous run?
  • SCRIPT: the path to reaction_box_no_slurm.py
• change the path to xTB in reaction_box_no_slurm.py

External Programs:

The procedure relies on xTB for the quantum chemical calculations and Slurm for job submission Also, a local version of xyz2mol (xyz2mol_local.py) is used to analyze the structures on the trajectory.

https://github.com/jensengroup/xyz2mol

Finally, the search is executed by:

./control_metadyn_runs.py reactant_smiles.csv

A folder named $S_FACTOR_$K_PUSH_$ALP is created, where the results are saved Results are saved in as dataframes in .pkl files: 1 for each run

The results for all runs can be combined using combine_runs.py, which collects all recorded reactions and extracts all 1-step reactions, both are saved as .csv files.

Files for the meta-dynamics product search can be found in metadyn_product_search

RMSD-PP procedure

If products are from the meta-dynamics product search, reactant and product structures can be extracted from the meta-dynamics search. This option is set by calling --use-structures when creating the submission file in rmsd_pp_no_slurm.py. Otherwise structures are embedded using RDKit.

Usage

A .csv file containing the reactions to get barrier estimates mus be created: these reactions can stem from either meta-dynamics search, systematic search or something else entirely. The .csv file must contain one row for each reaction with at least two columns:

• reactant_smiles_am containing the mapped SMILES of the reactant
• product_smiles_am containing the mapped SMILES of the product Then change parameters CPUS, MEM, MAX_QUEUE and SCRIPT in control_rmsd_paths.py
• CPUS: number of cpus each meta-dynamics job can use
• MEM: amount of memory each meta-dynamics job can use
• MAX_QUEUE: maximally allowed number of jobs to be submitted to Slurm at the same time
• SCRIPT: the path to rmsd_pp_no_slurm.py

for meta-dyanmics products

create_job_names.py can be used to make the reaction.csv file after meta-dynamics runs. By default 5 entries per reaction is created. RXN_INDEX_LIST must be changed to the list of reactant indexes used in the meta-dynamics search

The RMSD-PP paths are then computed by running

./control_rmsd_paths.py reactions.csv

External Programs:

The procedure relies on xTB for the quantum chemical calculations and Slurm for job submission Also, a local version of xyz2mol (xyz2mol_local.py) is used to analyze the structures on the trajectory.

https://github.com/jensengroup/xyz2mol

Results for each run is saved in a .pkl file named according to the job name.

The results for all runs can be collected by

./collect_dataframes.py reactions.csv

which returns an updated .csv file

Files for the RMSD-PP barrier estmation procedure is found in rmsd_pp_barrier_estimate

TS validation procedure

This procedures optimizes a TS based on the specified DFT functional and basis set and guess structures from the RMSD-PP procedure.

Usage

As input is needed the results from the RMSD-PP procedure: both the paths containing the TS guesses, and the updated .csv file containing barrier estimates for the reactions.

Change the parameters cpus, mem, max_queue and script in control_dft_ts.py to fit your requirements

• cpus: number of cpus each job should use
• mem: amount of memory each job should use
• max_queue: maximally allowed number of submitted jobs at a time
• script_path: path to run_dft_ts.py

remember to change the paths to xTB and Gaussian.

External Programs:

The procedure relies on Gaussian for the quantum chemical calculations and Slurm for job submission Also, a local version of xyz2mol (xyz2mol_local.py) is used to analyze the structures on the trajectory.

https://github.com/jensengroup/xyz2mol

The calculations are done by calling

./control_dft_ts.py rmsd_output.csv

Files for the meta-dynamics product search can be found in dft_ts_validation