Figure 1.
Overview of Spatial Integration for Imputation and Deconvolution (SIID). Given Xenium and Visium expression matrices AX and AV, respectively, with corresponding gene panels GX and GV (with GX ⊂ GV), and a Xenium to Visium spatial mapping Γ, SIID finds a NMF decomposition of a latent gene count matrix AU, with corresponding factorizations of AX and AV, into location to latent factor assignment matrix P and the gene expression matrix Q for latent factors. The estimated parameters are used to impute the absent gene expression in Xenium data and predict the cell type mixture proportions for each Visium spot.
