Python implementation of BoundLess Objective-free eXploration (BLOX)* for discovery of out-of-trend materials using Stein novelty.
* K. Terayama, M. Sumita, R. Tamura, D. T. Payne, M. K. Chahal, S. Ishihara, K. Tsuda, "Pushing property limits in materials discovery via boundless objective-free exploration," Chemical Science, 2020. [DOI: 10.1039/D0SC00982B]
- Python > 3.6
Download or clone the github repository, e.g., git clone https://github.com/tsudalab/BLOX
-
Preparation
- feature_list_of_observed_data.csv: Feature list of the observed materials. As an example, morgan fingerprints of 10 ZINC molecules are prepared.
- properties_of_observed_data.csv: List of the observed properties by experiment or simulation. The lengths of "feature_list_of_observed_data.csv" and "properties_of_observed_data.csv" should be the same. Here, we prepared three properties: 1 absorption wavelength, 2 intensity, and 3 molecular weight data. The absorption wavelength and intensity data were calculated by DFT simulations.
- feature_list_of_unchecked_data.csv: Feature list of the unchecked materials. Here, we prepared morgan fingerprints of 100 ZINC molecules.
- properties_of_unchecked_data.csv: Please note that these data are not prepared in the actual situation. Here, to demonstrate BLOX, we calculated absorption wavelength, intensity, and molecular weight data of the unchecked data in advance.
-
Exploration by BLOX
python explore_by_BLOX.py feature_list_of_observed_data.csv feature_list_of_unchecked_data.csv properties_of_observed_data.csv --prediction_model RF --iteration_num 20 --dimension 2- The plotted image of observed data and recommended data is saved in the fig directory.
- The prediction model can be set as '--prediction model RF' for random forest.
- The kernel parameter of the kernelized Stein discrepancy can be set as '--sigma XX' without '--adaptive' option.
- If the dimension is set as 2 (--dimension 2), the absorption wavelength and intensity properties are used for exploration. If the parameter is set as 3, all 3 properties are used.
-
Recommend the next candidate by Stein novelty
python recommend_next_data.py feature_list_of_observed_data.csv feature_list_of_unchecked_data.csv properties_of_observed_data.csv --dimension 2- The recommend candidate and its predicted properties are saved in "recommend_data_by_Stein_novelty.csv."