Figure 1.

Flowchart of the proposed BayesDeep: A. BayesDeep integrates the spot-resolution molecular profile $\mathit{Y}$ from NGS-based SRT data, the single-cell-resolution image profile $\mathit{X}$ from the paired AI-reconstructed histology image, and the spot-cell geospatial profile $\mathit{G}$ to recover gene expression at the single-cell resolution $\mathit{\Theta }$. B. The hierarchical formulation of the BayesDeep model is based on a Bayesian regularized negative binomial regression model with grouped observations. C. BayesDeep estimates the association between the single-cell-resolution molecular and image profiles $\mathit{B}$ and predicts the single-cell-resolution molecular profile $\mathit{\Theta }$. D. Several downstream analyses can be enhanced based on the availability of the single-cell-resolution molecular profile $\mathit{\Theta }$, including identifying distinct cell populations, elucidating the process of tumorigenesis via pseudotime analysis, and exploring the mechanisms of cell-cell communication.