This README documents the necessary steps to set up the running environment for Kalstra.

A Novel Hybrid Architecture Integrating KAN and Sequential-Transformer for Robust Cell Type Annotation in Cross-Species Transcriptomics

The code runs under Python 3.9.19 and Tensorflow 2.15.0. Create a Tensorflow environment and install required packages, such as "numpy", "pandas", "keras" and "scanpy" . Please refer to requirements.txt for more details.

pip install -r requirements.txt

• After setting up the above files, execute Python files sequentially in the folder KALSTRA.

• Training results are saved into modelsave/epoch200.txt

LSKAN.py: ATLSTM layer and CFKAN model
GMHA.py: Attention layer and BAFFN layer
model.py: Integrate LSKAN and GMHA into KALSTRA
training.py: Model training
testing.py: Prediction results of dataset

Have any questions or issues related to the repository, please contact Dr. Binhua Tang (bh.tang@hhu.edu.cn).