Fig. 2: Simulation experiments.
From: sCellST predicts single-cell gene expression from H& E images
a Simulation framework illustrating cell image and gene expression (GE) attribution under the cell scenario; other scenarios are presented in the SupFig.10. b–e Mean Pearson correlation for different sets of genes in the test dataset (error bars correspond to 95% confidence interval obtained via bootstrapping). The number of highly variable genes (HVG) is 1000 and the number of marker genes (MG) is 51. b Comparison across different scenarios (cell, centroid, random) at the bag and instance level on HVG. c Comparison across different sets of genes (HVG or MG) at the bag and instance level in the cell scenario. d Comparison across different learning frameworks (MIL or supervised) on different sets of genes (HVG and MG) in the cell scenario. e Comparison across different instance encodings (derived from SSL embeddings or one hot encoded cell types) on different sets of genes (HVG and MG) in the cell scenario.
