PF-OHIP

This manual contains the models implemented in the paper "D. Vijay Anand, Qiang Xu, JunJie Wee, Kelin Xia, Tze Chien Sum. Topological Feature Engineering for Machine Learning Based Halide Perovskite Materials Design. npj Computational Materials. 8, 203. (2022).".

Code Requirements

---

    Platform: Python>=3.6, MATLAB 2016B
    Python Packages needed: math, numpy>=1.19.5, scipy>=1.4.1, scikit-learn>=0.20.3, GUDHI 3.0.0, GeneralisedFormanRicci==0.3

Pipeline of PF-GBT Models

---

Feature Generation

Persistent Homology

Here, we provide two possible Feature extraction codes. One is written in python and one is written in MATLAB.

PrepareHOIPFeatures.m --> Extracts the betti numbers from the 30 atom sets barcode informations (MATLAB Feature Extraction)
PH_feature.py --> Generates the 30 atom set barcodes from file directory of all *.pdb files storing atom coordinates for OIHP structures. 

Persistent Forman Ricci Curvature

Note that the code for Persistent Forman Ricci curvature requires GUDHI 3.0.0 and GeneralisedFormanRicci v0.3.

FRC_feature.py --> Generates all the FRC values from file ph_input.npy storing atom coordinates for all the necessary atom sets in OIHP structures. 

Classification

Here, we provide the t-SNE codes for the classifications of 9 types of OIHP structures.

Persistent Homology

The last 100 frames for MAPbBr3 Cubic structures can be found in ./PH_OUT_MAPbBr3_Cubic_CNPbX/ folder. Both MATLAB and some python code for classification process. Some code snippets included were also used to generate Figure 3(a).

tSNE_Trunc_MAPbX.m --> Actual t-SNE classification code used for PH features. 
MAPbX3_classify_CHNPbX.py --> Additional t-SNE classification using just CHNPbX PH features. 
MAPbX3_classify_CNPbX.py --> Additional t-SNE classification using just CNPbX PH features. 

Persistent Forman Ricci Curvature

The features for MAPb3__CNXPb_data_feat.npy have been provided where = Br, Cl or I and = Cubic, Orthorhombic or Tetragonal.

FRC_alpha_gen_CNPbX.py --> The parallel processing code for generating all the FRC values from 4500 OIHP molecular dynamic simulation frames. 
FRC_alpha_CHNPbX.py --> Code to compute the statistical attributes or the FRC feature vectors from the OIHP structures (including hydrogen atoms).
FRC_alpha_CNPbX.py --> Code to compute the statistical attributes or the FRC feature vectors from the OIHP structures (excluding hydrogen atoms).
FRC_alpha_classify_CHNPbX.py --> t-SNE classification using just CHNPbX features. 
FRC_alpha_classify_CNPbX.py --> t-SNE classification using just CNPbX features.
FRC_alpha_classify_CNPbX_CHNPbX.py --> t-SNE classification using both CHNPbX and CNPbX features.

Bandgap Predictions

Here, we provide the codes for the bandgap prediction of the OIHP structures.

./TPD_ML_predictions/ --> Folder for the Traditional Perovskite Descriptors ML Model. 
./PH_ML_predictions/ --> Folder for the PH-based GBT Model. 
./FRC_ML_predictions/ --> Folder for the PFRC-based GBT Model.

Cite

If you use this code in your research, please considering cite our paper:

• D. Vijay Anand, Qiang Xu, JunJie Wee, Kelin Xia, Tze Chien Sum. Topological Feature Engineering for Machine Learning Based Halide Perovskite Materials Design. npj Computational Materials. 8, 203. (2022).