Figure 2. Multicellular factor analysis of a single-cell atlas of myocardial infarction.

(A) Simplified experimental design of a single-cell atlas of acute heart failure following myocardial infarction from Kuppe et al., 2022a. The lower panel shows the factor scores of the 27 samples inferred by the model. The condition and technical batch label of each sample are indicated next to each row. Samples are sorted based on hierarchical clustering. The middle panel shows the -log10 (adj. p-values, Kruskal-Wallis test) of testing for associations between the factor scores and the condition (myogenic: n = 13, ischemic: n = 9, fibrotic: n = 5) or batch label. The upper panel shows the percentage of explained variance of each cell-type expression matrix recovered by the factor. (B) Uniform Manifold Approximation and Projection (UMAP) embedding of the factor scores of each sample in the acute heart failure atlas. (C) Distribution of the scores of Factor 1 across different conditions. (D) Distribution of the scores of Factors 2 and 4 across different technical batches. Data information: In (A–C), myogenic: n = 13, ischemic: n = 9, fibrotic: n = 5. In (A, D), A: n=8, B: n=19. In (C, D) data is presented as box plots where the middle line corresponds to the median, the lower and upper hinges correspond to the first and third quartiles, and the whiskers extend no further than 1.5× interquartile range (IQR). Adjusted p-values from Wilcoxon test.

Figure 2—figure supplement 1. Estimation of a multicellular latent space of acute heart failure using multicellular factor analysis and scITD, and application of multicellular factor analysis for lupus samples.

(A) Multidimensional scaling (MDS) embedding of the multicellular factor scores of each sample in the acute heart failure atlas. (B) scITD Tucker decomposition of the acute heart failure atlas. The lower panel shows the factor scores of the 24 samples inferred by the scITD model. The condition (myogenic: n=13, ischemic: n=6, fibrotic: n=5) and technical batch labels (A: n=6, B: n=18) of each sample are indicated next to each row. Samples are sorted based on hierarchical clustering. The middle panel shows the -log10 (adj. p-values) testing for associations between the factor scores and the condition or batch label (analysis of variance - ANOVA). The upper panel shows the percentage of explained variance of each cell-type expression matrix recovered by the factor. (C) Distribution of silhouette widths of each sample modeled by multicellular factor analysis and scITD grouped by their condition or their technical label. Adjusted p-values <0.05 of Wilcoxon tests are shown. Number of samples across models and sample labels as in Figure 2—figure supplement 1B and Figure 2A. (D) Loading scores of multicellular factor analysis’ Factor 1 and scITD’s Factor 1 for POSTN and TTN across cell types. (E) Multicellular factor analysis of peripheral blood mononuclear cells from patient lupus samples, each before (n=8) and after interferon (IFN)-beta stimulation (n=8.) The lower panel shows the factor scores of the 16 samples inferred with multicellular factor analysis. The stimulation labels of each sample are indicated next to each row. Samples are sorted based on hierarchical clustering. The middle panel shows the -log10(adj. p-values, Kruskal-Wallis test) of testing for associations between the factor scores and the stimulation. The upper panel shows the percentage of explained variance of each cell-type expression matrix recovered by the factor. Data information: In C data is presented as box plots where the middle line corresponds to the median, the lower and upper hinges correspond to the first and third quartiles, and the whiskers extend no further than 1.5× interquartile range (IQR). Adjusted p-values from Wilcoxon test.