Extended Data Fig. 2: Quality assessment and processing of scRNA-seq data.
From: Rapid non-uniform adaptation to conformation-specific KRAS(G12C) inhibition
a, b, Gene counts as a function of UMI count. Cells are grouped by length of G12Ci treatment (a) or tumour model (b). c, The number of cells expressing a gene, as a function of its average count across the dataset. d, Variance as a function of mean expression. Technical variance (that is, variability attributed to technical factors) was calculated by the expression of ribosomal genes. n = 10,177 single cells in a–d. e, The per cent of variance explained by various experimental factors. A number of variables had a meaningful contribution to the variance of the dataset (that is, they accounted for greater than 1% of the variation), suggesting the need to correct for these potentially confounding factors in downstream analysis. f, Dimensionality reduction and covariate regression using the ZINB-WaVE algorithm. The K parameter of 2 was chosen, as this minimizes batch and other covariate effects. g, t-SNE projection showing single cells coloured by length of inhibitor treatment. h, Parameters used to cluster cells by using the Density Cluster algorithm. i, Cluster distribution in the indicated projections (top) and cell line composition of each cluster (bottom), showing a similar representation of cells from different tumour models in each cluster. j, Silhouette-width analysis to assess the appropriateness of clustering. Negative values indicate cells that have been inappropriately assigned. k, t-SNE projection of KRAS(G12C) single cells with the three inhibitory trajectories identified by the Slingshot algorithm.
