Fig. 2: SPLISOSM produces well-calibrated and permutation-free P values in simulation.
From: Mapping isoforms and regulatory mechanisms from spatial transcriptomics data with SPLISOSM
a, Six simulation scenarios with SV at different layers. Regional variability was generated from two binary and two continuous spatial covariates Iso, isoform. b, Q–Q plots of SV test P values across 1,000 simulated genes per scenario. Regional gene expression (scenarios 4–6) was included to introduce artifacts in the observed ratios. c, Q–Q plots of DU test P values aggregated from 2,000 tests (1,000 genes × 2 covariates) per covariate type. Variable isoform usage independent of covariates (scenarios 2 and 5) was sampled from a Gaussian process (GP) to mimic spurious differential associations. For binary covariates, two-sided t-tests were performed on individual isoforms with P values combined at the gene level using Fisher’s method. For GLM and GLMM, null models with zero effect size were fitted and tested using the score statistic. Red text indicates components with spatial variability.
