Supplementary Figure 10: Assessing statistical calibration of SpatialDE through simulations.

(A-B) Simulation of bell curve shaped data on the mouse olfactory bulb coordinates. (A) Example of six bell curves with different radii. (B) Results from applying SpatialDE to data from 3,000 bell curves with different radii and stratified over different levels of simulated noise (fraction of spatial variance, FSV). Red line denotes P=0.05 significance level, vertical blue dotted lines indicate smallest and largest pairwise distances observed in mouse olfactory bulb data respectively. Black dots denote negative log P-values (left axis), while blue dashes indicating statistical power for detecting true simulated SV genes (fraction true positives) for each bell radius (right axis). (C) Analogous results as in B, however when simulating data from the generative model underlying SpatialDE, considering different values for the fraction of spatial variance (FSV) and length scales, considering 3,000 simulations. (D) Scatter plot of inferred length scales (y-axis) versus simulated length scale (x-axis) for the data shown in C. (E) Simulation of 3,000 genes from the null model with no spatial covariance. Bars denote the fraction of (false positive) SV genes for different SpatialDE P-value thresholds (x-axis). The proportion of false positive genes was lower than the controlled family-wise error rate (FWER), indicating that the test is conservative.