Fig. 1

Most spatial transcriptomics datasets contain an excessive proportion of zero values. A The majority of spatial transcriptomics datasets (x-axis) contain a substantial proportion of zero values (y-axis). B The proportion of zero values in each dataset (y-axis; logit-transformed) is negatively correlated with the total counts per location (x-axis). Tomo-seq (zero proportion = 0.797; total count per location = 0.198 million) is an outlier and is not displayed on the panel. C The proportion of zero values for each gene (y-axis) is plotted against the expression mean (x-axis) for three example datasets that include seqFISH+, Slide-seqV2, and 10x_MB(C). The zero vs mean trend is fitted by either a Poisson model (blue line) or a negative binomial model (red line). D Mean square error (MSE) for the estimated zero proportion (y-axis; log transformed) based on either the Poisson model (blue line) or the negative binomial (red line) across datasets (x-axis). In both A and D, the gray dotted line separates smFISH-based technologies from the sequencing-based technologies and solid line separates single-cell resolution technologies to spot-level resolution technologies