Fig. 2
From: scMET: Bayesian modeling of DNA methylation heterogeneity at single-cell resolution
Feature selection using scMET characterises cellular heterogeneity on the mouse frontal cortex. a UMAP [26] representation of neuron sub-populations present in the mouse frontal cortex dataset by combining the top 4000 HVFs identified by scMET across distal H3K27ac and H3K4me1 genomic contexts. Cells are colored according to the original cell sub-population assignments obtained by Luo et al. b Clustering performance in terms of adjusted Rand index (ARI), for varying number of HVFs. HVF selection was based on scMET’s residual overdispersion parameters εj (yellow), binomial variance (gray), Gaussian variance (cyan), normalized dispersion (NormDisp, red), and random selection (blue). A finite grid of HVFs was used for ARI evaluation and non-parametric regression was used to obtain a smoothed interpolation across all values (Methods). c Identifying HVFs for the distal H3K27ac genomic context. Red points correspond to features being called as HVF (EFDR = 10% and percentile threshold δE = 90%). To ease interpretation, each element is linked to its nearest gene. Labels highlight known neuron marker genes that were used in the [9] study to define the different cell populations. d Example distal H3K27ac HVFs whose methylation patterns distinguish the two broad neuronal populations. Panel titles correspond to the nearest genes