multi-fidelity Bayesian optimization of covalent organic frameworks

🚀 this repo contains data and code to reproduce the results for:

N. Gantzler, A. Deshwal, J. Doppa, C. Simon. "Multi-fidelity Bayesian Optimization of Covalent Organic Frameworks for Xenon/Krypton Separations" (2023) ChemRxiv link

we describe the sequence of steps we took to make our paper reproducible. the output of each step is saved as a file, so you can start at any step.

required software

required software/packages:

• Python 3 version 3.8 or newer (for MFBO)
• Julia version 1.7.3 or newer (for molecular simulations and assembling data)
• Zeo++ (for computing structural features of the COFs)

the COF crystal structures

we obtained the dataset of the COF crystal structure files (.cif) from Materials Cloud and stored them in data/crystals. see here for the COF naming convention.

the top COF 19440N2: the paper associated with the COF exhibiting the largest high-fidelity Xe/Kr selectivity is:

"J. Am. Chem. Soc., 2019, 141, 16810-16816", 10.1021/jacs.9b07644, Unveiling Electronic Properties in Metal–Phthalocyanine-Based Pyrazine-Linked Conjugated Two-Dimensional Covalent Organic Frameworks they report on two novel COFs, one with Zn and one with Cu. the COF with Cu is the 19440N2 (top COF), and the Zn COF is 19441N2.

molecular simulations of Xe/Kr adsorption in the COFs

we performed mixture grand-canonical Monte Carlo simulations to predict the adsorbtion properties of the COFs by running htc_screening/sbatch_submit_script.sh on a HPC cluster which uses SLURM. the molecular simulation code in Julia is contained in htc_screening/multi_fidelity_simulation_script.jl. this script relies on the PorousMaterials.jl package in Julia and runs both low- and high-fidelity simulations.

• the raw simulation output files (.jld's) are in data/simulations/
• the simulation data is organized as .csv in targets/{gcmc_simulation.csv, henry_calculations.csv}

we employed PorousMaterials.jl v0.4.2 for the mixture GCMC and Henry coefficient calculations.

COF descriptors

structural/geometrical descriptors

we computed structural descriptors using Zeo++ by running the script descriptors/submit_zeo_calculations.sh, which runs locally and computes descriptors for all of the COFs. (if running on a HPC cluster with SLURM, see descriptors/submit_slurm_job.sh).

• the raw Zeo++ outputs per-crystal are stored in descriptors/zeo_outputs/.
• the raw Zeo++ output for all crystals is compiled in the three files descriptors/summary_*.
• the relevant features of the COFs from Zeo++ are assembled in descriptors/geometric_properties.csv

compositional descriptors

we computed the compositional descriptors of the COFs using PorousMaterials.jl (version 0.4.2 or newer) by running descriptors/cof_features.jl.

• we stored them in descriptors/chemical_properties.csv

joined structural and compositional features

the COF descriptors are summarized in descriptors/cof_descriptors.csv, which joins descriptors/geometric_properties.csv and descriptors/chemical_properties.csv.

amalgamating the data for MFBO

we read in and amalgamated the structural and compositional features and low- and high-fidelity simulation results into targets_and_raw_features.jld2 by running the notebook Prepare_Data_and_Preliminary_Analysis.ipynb. this .jld2 file is what we read into our Jupyter notebooks for Bayes Opt. the data are conveniently stored as a dictionary of arrays. the outputted file is present in run_BO.

Bayes Opt

initialization

we generate the list of initializing COF IDs using the run_BO/generate_initializing_cof_ids.ipynb. this writes a file search_results/initializing_cof_ids_normalized.pkl that we read into the Bayes Opt notebooks.

single- and multi-fidelity Bayes Opt

finally, the two notebooks:

• run_BO/MultiFidelity_BO.ipynb
• run_BO/SingleFidelity_BO.ipynb contain the Python code for running Bayes Opt.

the results from each run are stored in search_results to be read into our figs/viz.ipynb notebook next for analysis.

visualizations/analysis

we draw plots to summarize the Bayes Opt search results results using viz.ipynb. the outputted figures are stored in the figs directory.

overview of directories

• data: contains simulation input and output files.
• htc_screening: code to run and analyze the molecular simulations of adsorption in the COFs
• benchmarking_sims: contains code and analysis to determine the number of cycles required to reduce statistical error for GCMC simulations and Henry Coefficients below a given threshold.
• descriptors: contains the scritps used to generate the descriptors of the COFs
• figs: for creating the figures in our paper
• search_results: the BO search results organized by the type of normalization scheme used and subdivided by the type of BO search carried out. also contains files for sets of initializing COFs.
• targets: contains the high-fidelity GCMC simulation results and Henry coefficient calculation results for each material in the study as CSV files.