Figure 2.

Evaluation results on seqFISH dataset

(A) Observed cell-type distributions over spatial locations.

(B) Example well-predicted genes: (top) observed, (bottom) imputed. M.I., Moran’s I statistics; C.S., cosine similarity score.

(C) Example less-well-predicted genes: (top) observed, (bottom) imputed.

(D) Boxplots of cosine similarity scores (CSSs) for all methods. The first box shows the statistics of a subset of genes, with predicted uncertainty below the median of all genes.

(E) Line plot of CSSs aggregated in different quantile ranges. The x axis denotes the quantiles of predicted uncertainty, the y axis denotes the median CSS of genes below the corresponding uncertainty quantile.

(F) Bar plot for proportions of unprobed marker genes in each cell type.

(G) The top 1 ranked marker gene for each cell type. The suffix “imp” indicates unprobed genes imputed from the single-cell reference.

(H) Spatial pattern plots of an example spatial ligand-receptor interaction pair: (top) on observed data, (bottom) on imputed data. In each row, the leftmost plot shows the probabilities of spatial interaction over spots, while the remaining two show the expression patterns of the involved genes.

(I) Venn diagram of significant ligand-receptor pairs in observed and imputed ST expression matrices.

(J) Plots of two spatial ligand-receptor pairs (top and bottom) of unprobed genes.