Fig. 3

Classification of cassette exons based on single-cell splicing patterns in iPS cells. a Single-cell splicing rate (PSI) distributions of the 5 splicing categories (inspired by Song et al. [12]) in 84 iPS cells. Intermediate splicing categories that can only be defined based on single-cell information are framed by a gray box. b Variation of PSI (standard deviation) across cells as a function of the average inclusion rate of cassette exons across 84 iPS cells, colored according to their respective splicing category as defined in a. The solid black line denotes the LOESS fit across all cassette exons. c Performance of logistic regression models for predicting splicing categories based on genomic features. Shown is the receiver operating characteristics for each splicing category and the macro-average (area under the curve, AUC). d Prediction performance of alternative regression models for each splicing category, either considering a model trained using genomic features (“genomic,” left), genomic and all DNA methylation features (“genomic and methylation,” center) as well as only DNA methylation features (“methylation,” right). The genomic model includes k-mers, conservation scores, and region lengths (see Fig. 1c). The genomic and methylation model additionally includes DNA methylation features. The methylation model includes average DNA methylation features per sequence context. Splicing categories are coded in color as in a. Error bars denote ± 1 standard deviation across 4 repeat experiments. e Distribution of DNA methylation levels in the upstream exon (C1) per splicing category. Methylation is decreased in underdispersed exons