Skip to main content
. 2022 Apr 25;3(4):505–517. doi: 10.1038/s43018-022-00356-3

Extended Data Fig. 3. Comparison between BayesPrism and other deconvolution methods on pseudo-bulks across different sequencing platforms and biological samples.

Extended Data Fig. 3

Boxplots show the cell type-level Pearson’s correlation coefficient and MSE for the deconvolutions of pseudo-bulk human PBMC scRNA-seq (a and b, n=200 pseudo-bulks simulated using 253 single cells collected from one healthy adult) and mouse cortex single nucleus-seq (c and d, n=200 pseudo-bulks simulated using 295 single cells collected from one mouse). Boxes mark the 25th percentile (bottom of box), median (central bar), and 75th percentile (top of box). Whiskers represent extreme values within 1.5 fold of the inter quartile range. One-sided p values were shown for cell type fractions inferred by BayesPrism (updated θ using the marker free mode) and those by the second-best methods ranked by the median value. T test was used for MSE and z test was performed on Fisher’s Z-transformed cell type-level correlation coefficients (see Methods).

Source data