Fig. 2: Percentile correlations and isoform clustering.
From: acorde unravels functionally interpretable networks of isoform co-usage from single cell data
a Percentile correlation algorithm. For each isoform, cell type-level expression is summarized using percentiles (0–10) as a proxy of the isoform’s expression distribution in each of the cell types. Then, Pearson correlations are computed using the percentile-summarized expression of all isoforms, obtaining a percentile correlation matrix. b Correlation density distributions. Pairwise isoform correlations were computed using Pearson, Spearman, and percentile+Pearson correlation. c Clustering pipeline. The percentile correlation matrix is first used as a distance matrix for hierarchical clustering. After dynamic cluster generation, noisy clusters are refined by a three-step semi-automated process. d Clusters generated after applying the acorde clustering pipeline to the mouse neural dataset. Cell-level mean expression (scaled, see “Methods”) is computed for all transcripts and then aggregated as the global cell type mean, represented by the red line. Gray area corresponds to cell type mean ± standard deviation. Astr: astrocytes, End: endothelial cells, GABA: GABA-ergic neurons, Glut: glutamatergic neurons, Micr: microglia, Oligo: oligodendrocytes, OPC: oligodendrocyte precursor cells.
