Fig. 3

Benchmarking SRTsim against six existing scRNAseq simulators for generating SRT data based on the CosMx SMI reference data. A Violin plots show the distributions of six metrics (six panels) in the reference data (real; x-axis) and in the synthetic data generated by different simulators (x-axis). The six metrics include four gene-wise metrics (expression mean; variance; coefficient of variation, cv; and zero proportion) and two location-wise metrics (zero proportion and library size). The examined simulators include SRTsim, scDesign2, scDesign2 without copula (scDesign2 (ind)), ZINB-WaVE, SymSim, and three variants of the splatter package (splat simple, and splat). SPARsim and kersplat were unable to generate the synthetic data for this sample, therefore were excluded from the comparison. Results are shown for the reference sample Lung12 FOV1. Results for the other four reference samples are provided in the supplemental figures. B The spatial expression patterns of three representative genes (IGHA1, KRT19, SLPI) are displayed in the reference data and in the synthetic data generated by SRTsim, scDesign2, and ZINB-WaVE. C The heatmap shows the proportion of the preferred count model in each dataset. The right color bar represents the SRT techniques: 10X (orange), Spatial Transcriptomics (dark green), DBiT-seq(pink), seqFISH+ (cyan), STARmap (purple), SlideseqV2 (yellow), CosMx SMI (red), and MERFISH (blue). The corresponding dataset names are listed in Table S1. NB, negative binomial; ZIP, zero-inflated Poisson; ZINB, zero-inflated negative binomial