Supplementary Figure 7: Validation and performance of cellAlign on non-imputed single-cell RNA sequencing data.
From: Alignment of single-cell trajectories to compare cellular expression dynamics

(a-b) Correlation between pseudo-time scores obtained from the normalized data (y-axes) versus those obtained from the imputed data (x-axes) for LPS (a) and PAM (b) trajectories. Scaled interpolated expression of the different 12 gene-modules calculated either using the normalized or imputed data are displayed beside and below the axes, respectively. Names of the different gene-modules, as appear in the original publication, are placed along a diagonal path to the left and below expression dynamics. Spearman correlation values between the imputed and non-imputed expression values appear on top. (c-f) Non-imputed, scaled expression (z-scores) of genes partitioned in the modules used for either global (c, d) or local (e, f) alignment under either LPS (c, e) or PAM (d, f) stimulations. Genes are ordered along the rows, while single cells are ordered by their pseudo-time scores on columns. Collection times and pseudo-time scores per single cell are displayed at the top. White matrix elements correspond to missing values (zeros). (g) Overall, non-imputed, scaled expression of a gene module by single cells (colors denote collection time) and interpolated points (black) along PAM (top) and LPS (bottom) trajectories. Values correspond to the 0.6 quantile scaled expression across the genes of the module. Dashed lines are point alignments from cellAlign global alignment. (h) Dissimilarity matrix and cellAlign global alignment of the non-imputed expression dynamics of one gene module along LPS (left to right) and PAM (top to bottom) trajectories. Histograms depict the distributions of collection time-points along either trajectory. Absolute values and signs of the pseudo-time shift elements, as derived from the alignment, are displayed as color bars at bottom. (i) Dissimilarity matrix and local alignment between non-imputed expression dynamics of a single gene module along LPS (left to right) and PAM (top to bottom) trajectories under similarity threshold of 0.25 of the range of values in the dissimilarity matrix. (j) Overall, non-imputed, expression of a gene-module whose expression pattern is conserved early following both LPS and PAM stimuli and deviates afterwards. Values correspond to the median interpolated scaled expression across module genes. Locally aligned points are red, whereas unaligned, non-conserved points in the trajectory are gray. (k) Distributions of the single-gene correlation difference between the aligned and un-aligned trajectories for global (pink, two-sided t-test p-value = 3*10−9, n = 67 genes) and local (top, two sided t-test p-value=0.027, n = 60 genes) alignment. Only those genes with less than overall 200 missing values in both LPS and PAM stimulated single cells participated in this analysis.