BiGATAE: a bipartite graph attention auto-encoder enhancing spatial domain identification from single-slice to multi-slices

We show how to operate on the Human dataset in BiGATAE/Tutorial.ipynb, and the operations on other datasets are in BiGATAE/train. We also store the raw data, intermediate results and final results of BC data and cSCC data in BiGATAE/tmp_data. The Human_DLPFC dataset is too large, please refer to Tutorial to get final results. BiGATAE is a data augmentation method. After getting the enhanced data, it can be used for various spatial clustering methods, such as stLearn, SEDR.

1. Human dorsolateral prefrontal cortex dataset
2. Breast cancer dataset
3. Human skin squamous cell carcinoma dataset

To execute the code, make sure to install the required packages listed below.

• python==3.8
• torch>=1.8.0
• scikit-learn==1.1.1
• matplotlib==3.4.2
• pandas==1.4.2
• numpy==1.22.3
• scanpy==1.9.1
• R==4.0.3