Figure 1.

Conceptual model. (A) Cell state–associated exons change across the phenotypic landscape of a single-cell population. Cell state–independent exons do not change across the phenotypic landscape. Low capture efficiency in scRNA-seq experiments adds additional technical variance depending on the number of captured mRNA molecules. The probability of capturing each alternative isoform depends on the underlying distribution of exons in the single-cell population. (B) Psix compares the likelihood of each single-cell observation given two models: model 1, in which the exon is cell state associated (probability of the cell's $\hat{\mathrm{\Psi }}$ given the average $\hat{\mathrm{\Psi }}$ of its k nearest neighbors), versus model 0, in which the exon is cell state independent (probability of the cell's $\hat{\mathrm{\Psi }}$ given the average Ψ of all cells in the data set). Model 1 is more likely for a cell state–associated exon. For a cell state–independent exon, the expected $\hat{\mathrm{\Psi }}$ of any cell is the same irrespective of its position in the cell state manifold. As a result, the expected value of the average $\overline{\mathrm{\Psi }}$ of a neighborhood of cells is the same as the global average $\overline{\mathrm{\Psi }}$. For this reason, the likelihood of model 1 is similar to the likelihood of model 0 for a cell state–independent exon.