Skip to main content
. 2022 Dec 28;11:e80380. doi: 10.7554/eLife.80380

Figure 5. Euclidean distance-to-centroid methods are unable to distinguish bona fide transcriptional noise from alternative cell fate specification.

(A) An increasing number of alveolar macrophage subclusters (as obtained with Leiden) are detected in three donors (aged 46, 51, and 75 years) from the Travaglini et al., 2020 (human lung cell atlas) dataset. (B–C) The new cell clusters are characterized by differential surfactant protein gene expression levels, as clearly seen on the uniform manifold approximation and projection (B) and dotplot (C) representations. (D–E) Transcriptional noise measurements, using the Euclidean distance to cell type mean (D) and 1 ‍ − membership using Scallop (E), demonstrate that only the latter method is able to distinguish bona fide transcriptional noise from the formation of new clusters that are transcriptionally stable.

Figure 5.

Figure 5—figure supplement 1. Expression of surfactant protein genes by human alveolar macrophages.

Figure 5—figure supplement 1.

Differential expression by alveolar macrophage cell clusters of the genes coding for surfactant proteins SFTPA1, SFTPA2, SFTPB, SFTPC, and SFTPD is shown for three donors (aged 46, 51, and 75) of the Travaglini et al. (human lung cell atlas) dataset.
Figure 5—figure supplement 2. Alveolar macrophages are the most affected by aging.

Figure 5—figure supplement 2.

Transcriptional noise in aged pancreatic β-cells. β-cells from young and old donors were isolated from the human aging pancreas dataset by Enge et al., 2017. (A) Uniform manifold approximation and projections (UMAPs) showing all pancreatic cell types and the selected β-cells with their corresponding cell type, age, and donor annotations. The β-cells from the 21-year-old donor are highlighted on the UMAP to point out that most cells from that donor appear clustered together on the UMAP, while the cells coming from the rest of the donors are spread over the whole plot. (B) Four methods for the quantification of transcriptional noise that yield cell-wise measurements: External RNA Controls Consortium (ERCC)-based (biological over technical variation), euc_dist (whole transcriptome-based Euclidean distance to cell type mean expression), euc_dist_invar (invariant gene-based Euclidean distance to average tissue expression), note that in this case the dataset only contains a single cell type, so the distance is computed to the average β-cell expression, scallop (1 - membership to most frequently assigned cluster). (C) Global Coordination Level per age category (number of divisions: 50) and Euclidean distance between 1000 randomly selected cells. (D) UMAPs showing the noise measurements per cell type using the same four methods (as in B). (E) Boxplots showing the distribution of transcriptional noise values for each donor. We can see that, rather than observing an age-dependent pattern, the 21-year-old donor presents much lower transcriptional noise values than the rest of the donors. However, there is no significant difference between the 22-year-old donor and the rest of the donors. We conclude that the statistically significant differences between the young and old age categories can be attributed to the abnormal noise values obtained for the 21-year-old donor. (F) UMAP and boxplots showing insulin expression per cell and per donor. The 21-year-old donor presents an abnormally low insulin expression.