STIFT enables batch effect removal, spatial domain identification, and exploration of developmental dynamics during developmental and regenerative processes.
STIFT first uses developmental spatiotemporal optimal transport to establish probabilistic mappings between spots across consecutive time points using the gene expression information and spatial coordinates information of all slices. Second, it simultaneously constructs a spatial neighbor network using the spatial coordinates information within each slice and a temporal relation network from the probabilistic mappings. Third, it integrates these two networks to construct a spatiotemporal graph. Finally, it takes the spatiotemporal graph and gene expression information to Graph Attention Autoencoder (GATE) with triplet learning informed by temporal relations to generate integrated embeddings that preserve both spatial organization and developmental trajectories.
git clone https://github.com/TheLittleJimmy/STIFT.git
cd STIFT
#create an environment called env_STAligner
conda create -n env_STIFT python=3.9.19
#activate your environment
conda activate env_STIFT
pip install -r requirements.txt
pip install .
Tutorial provides the basic workflow of STIFT.
If there are any questions, please contact the author at qiji@link.cuhk.edu.hk. The author is happy to help!