Spatial transcriptomics (ST) can provide vital insights into tissue function with the spatial organization of cell types. However, most technologies have limited spatial resolution, i.e., each measured location contains a mixture of cells, which only quantify the average expression level across many cells in the location. Recently developed algorithms show the promise to overcome these challenges by integrating single-cell and spatial data. In this review, we summarize spatial transcriptomic technologies and efforts at cell-type deconvolution. Importantly, we propose a unified probabilistic framework, integrating the details of the ST data generation process and the gene expression process simultaneously for modeling and inferring spatial transcriptomic data.