This repo provides codes and scripts in which generate descriptors and predict Henry's constants and the heats of adsorption for a given set of MOFs and adsorbates[1]. There are three directories: descriptor_calculation, classification, and ensemble_modeling.
This directory contains source codes of generating AP-RDF and molecular descriptors that are used in this study.
This directory serves as a source code for Section 3.I. Data_S1.csv is a train/test set consisting of 471 MOFs and 30 molecules, while Data_S2.csv is a hold-out set with the same MOFs but new 15 molecules. Data_S2.csv is never faced during the classification training procedure and is for assessing generalizability of our scheme throughout this work.
This directory is a source code for Section 3.II. set_1.csv and set_2.csv are the result of Section 3.I. from Data_S1.csv and Data_S2.csv, respectively. set_1.csv contains 12,960 MOF-molecule pairs and set_2.csv has 7,065 pairs.
To run the ensemble model, type the following command on Terminal: python pipeline.py set_1.csv set_2.csv
This will create one subdirectory: data
data directory consists of:
- Predicted results for the training and test sets in each data split (numbered directories correspond to the random seeds used during the data split)
- Predicted results for set 2
[1] X. Yu, S. Choi, D. Tang, A. J. Medford, D. Sholl, Efficient Models For Predicting Temperature-dependent Henry’s Constants and Adsorption Selectivities for Diverse Collections of Molecules in Metal-Organic Frameworks (Submitted)
- NumPy
- SciPy
- Pandas
- scikit-learn