Deep Powder Crystallography

Source code to accompany our upcoming paper, Towards End-to-End Structure Determination from X-Ray Diffraction Data Using Deep Learning by Gabe Guo, Judah Goldfeder, Ling Lan, Aniv Ray, Albert Hanming Yang, Boyuan Chen, Simon JL Billinge, Hod Lipson.

Project Objective

Given x-ray diffraction pattern and partial chemical composition information, reconstruct 3D electron density function. AKA crystallography.

Reproducibility

Important: Will need to change all datapaths starting in /home/gabeguo/ to the corresponding filepaths on your system.

Environment

Python 3.10.10

pip install -r requirements.txt

Dataset

Source

Data can be downloaded from the Materials Project.

Also need to change API_KEY in download_charge_densities.py.

cd download_data
bash download_data.sh

Obtain Stable Crystals

cd download_data
bash record_stable_chemical_info.sh

Data Split to Train, Val, Test

You may edit the seed argument if you like, in run_split_dataset.sh.

cd download_data
bash run_split_dataset.sh

Note: Our own attempts to replicate the study indicate that the Materials Project dataset adds and removes some materials over time — thus, results may vary slightly from what is listed in the paper. For reproducibility, we supply the mpids of all the crystals used in training, validation, and testing in data_split_reproducibility_info.

Get Formula, Crystal System, and Spacegroup Info

cd download_data
bash record_chemical_aux_info.sh

Remove Duplicate Materials

cd download_data
bash remove_duplicate_materials.sh

Training Model

cd multiXRD_withFormula/new_experimental_scripts
CUDA_VISIBLE_DEVICES=x bash train_base_model.sh
CUDA_VISIBLE_DEVICES=x bash train_formula_ablation.sh

Testing Model

cd multiXRD_withFormula/new_experimental_scripts
CUDA_VISIBLE_DEVICES=x bash test_base_model.sh
CUDA_VISIBLE_DEVICES=x bash test_formula_ablation.sh