STGIC: A graph and image convolution-based method for spatial transcriptomic clustering
Fig 3
Workflow of the pre-clustering method AGC in STGIC.
Pixel coordinates of spots is used to compute the Euclidean distance based on which adjacency is constructed and then the pre-defined kernel function can be calculated. The top 50 PCs of each spot serve as the beginning feature input and are aggregated with the guidance of the kernel function to obtain spots’ embeddings for spectral clustering to generate a cluster assignment and the intra-cluster distance is subsequently calculated during each iteration. When the intra-cluster distance starts to increase, iteration terminates and the cluster assignment in the iteration presenting the minimum intra-cluster distance is adopted as the pre-clustering result.