Fig. 1: Overview of spatial omics scope (soScope) and its applications.

a The soScope framework. soScope integrates molecular profiles (\({{{\boldsymbol{X}}}}\)), spatial neighboring relations (\({{{\boldsymbol{A}}}}\)), and morphological image features (\({{{\boldsymbol{Y}}}}\)) from the same tissue using a unified generative model to enhance spatial resolution and refine data quality for diverse spatial omics profiles. b The probabilistic graphical model representation of soScope. Each of the \(N\) spots in the spatial data is considered an aggregation of \(K\) subspots at a higher spatial resolution. The subspot omics profile \(\widehat{{{{\boldsymbol{X}}}}}\) depends on both the latent states \({{{\boldsymbol{Z}}}}\) at the spot level and image features \({{{\boldsymbol{Y}}}}\) at the subspot level. The observed profile \({{{\boldsymbol{X}}}}\) is obtained by summing profiles from its \(K\) subspots. c The neural network structure in soScope. Original spatial profiles (\({{{\boldsymbol{X}}}}\)) and their spatial relations (\({{{\boldsymbol{A}}}}\)) are integrated using a graph encoder to infer the latent states of spots (\({{{\boldsymbol{Z}}}}\)). These latent representations are then combined with subspot-level image features (\({{{\boldsymbol{Y}}}}\)) to jointly learn likelihood parameters for modeling omics profiles at an enhanced resolution. The choice of probabilistic distribution is tailored to the specific omics type to reflect the data characteristics. d An overview of soScope experiments in this study. soScope is evaluated across multiple spatial omics platforms and diverse biological systems. Furthermore, soScope is extended to emerging spatial multiomics technology with simultaneous profiling of multiple molecule types.