Extended Data Fig. 3: Bulk RNA-Seq deconvolution and comparison of automatic and manual annotations in the COVID-19 lung cell atlas.
From: COVID-19 tissue atlases reveal SARS-CoV-2 pathology and cellular targets
a, b, Deconvolution of bulk RNA-Seq libraries from adjacent lung tissue. a, Mean proportion (y axis, error bars = s.d. estimates from bulk RNA-Seq deconvolution (hatched bars; from MuSiC55) and from sc/snRNA-Seq (filled bars) for each of 11 cell subsets (x axis) in each of 16 bulk RNA-Seq lung samples (panels) from 10 random samples of 10,000 cells each. b, Robustness of cell proportion estimates to the number of single cells sampled for the reference data. Mean proportion (y axis, from MuSiC) estimates for each of 11 cell subsets (colour dots) in each of 16 bulk RNA-Seq lung samples (panels) when using three independent samples of 1,000–10,000 cells from the single-cell reference (x axis). c–e, Agreement between automated and manual annotations. c. High consistency between automatic and manual annotations. The proportion (colour intensity) and number (dot size) of cells with a given predicted annotation (rows) in each manual annotation category (columns). d, e, UMAP embedding of myeloid (k = 24,417 cells or nuclei) (d) and T and NK (k = 9,950 cells); (e) cell profiles coloured by manually annotated subclusters (left) or automated predictions (right).
