Fig. 1

Overview of the Smoother framework. Smoother is a versatile and modular framework designed to incorporate spatial dependencies into various omics data analysis applications. The process initiates with the construction of a weighted spatial graph, derived from physical positions, histology, and additional features, which serves to represent spatial dependencies a priori (top). The prior is subsequently employed as a sparse loss function to regularize spatial variables, such as gene activities, cell-type compositions, and latent embeddings (bottom). Owing to its modular design, the spatial loss can be appended to preexisting models that were initially developed for non-spatial data, potentially bridging the gap between single-cell and spatial data analysis. The Smoother toolbox includes a selection of spatially aware versions of non-spatial models, including NNLS, DWLS, and SVR for cell-type deconvolution, and PCA, SCVI, and SCANVI for dimensionality reduction