Fig. 6: Structures enable better delineation of cellular identity.

a, Scatter plots showing the MOFA clustering using (1) RNA expression only (top left), (2) RNA expression and alternative splicing (bottom left) and (3) RNA expression, alternative splicing and structural heterogeneity (top right). Bottom right: ARI for the MOFA clustering results using different datasets as input. b, A heatmap showing reactivity of stage-specific structures across neuronal differentiation. Each row shows the reactivity of a 10 nt window, and each column is the reactivity for a single cell. Color scale: z-score transformed reactivity by row. Selected genes with structure-changing windows were labeled at the right of the heatmap. c, Top: pseudobulk reactivity of RPL41 in each stage of neuronal differentiation. The red dashed boxes indicate two structure-changing windows during neuronal differentiation. Single-nucleotide reactivity was smoothed using a 10 nt sliding window. Bottom: single-nucleotide reactivity is shown in the zoomed-in versions of the structure-changing regions. The blue curves were the average reactivity of cells in each stage. The light-blue shading indicates the standard error for each nucleotide. The cell number for each time point is labeled. d, Unsupervised clustering of RPL41 RNA structures in single cells during the neuronal differentiation process. The bar plot shows the number of cells that contain cluster 1 and cluster 2 in hESCs and different stages of neuronal differentiation. e,f, T-distributed stochastic neighbor embedding (TSNE) plot showing the distribution of reactivity of RPL41:160 (e) and RPL41:310 (f) in all single hESC and differentiated single cells.