Fig. 1: Overview of stClinic.

a stClinic integrates multi-slice omics data from the same tissue or different tissues, as well as multi-omics data from the same slice or different slices/technologies. Vertical integration aligns and integrates multi-omics data within the same slice, while diagonal integration does so across different slices. b Given multi-slice omics profiles (\({{\bf{X}}}\)) and spatial location (\({{\bf{S}}}\)) data as input, stClinic learns batch-corrected latent features (\({{\bf{z}}}\)) using a dynamically evolving graph, guided by a Mixture-of-Gaussians prior through Kullback-Leibler (KL) divergence regularization. c Given \({{\bf{z}}}\) and clinical data (\({{\bf{Y}}}\)) as input, stClinic quantifies the weights (\({{{\bf{W}}}}^{T}\)) of each cluster in clinical outcome prediction by representing each slice using a niche vector characterized by six geometric statistical measures relative to the population. d Integrated features \({{\bf{z}}}\) and weights (\({{{\bf{W}}}}^{T}\)) serve various purposes in dissecting niches from heterogenous tissues: identifying shared and condition-specific niches, assessing niche importance in phenotype prediction, transferring labels from the reference through zero-shot learning, and annotating labels across different types of omics datasets.