Fig. 1.

Illustration of BRIE2. a Reads are counted into isoform 1, isoform 2 or ambiguous groups according to its alignment identity, which constitutes a cell-by-gene-by-3 tensor. b The posterior distribution of isoform proportion PSI is defined by combining the likelihood from read counts and informative prior predicted by cell-level covariates and/or gene sequence features. c A logit- normal variational posterior and coefficients on covariates are optimised to approximate the exact posterior, where the evidence lower bound (ELBO) gain between including and excluding a certain cell feature set can be leveraged to select splicing phenotypes. d The selected differential splicing events or differential momentum genes on RNA velocity can be used as markers for downstream analysis, and the estimated PSI can be used for dimension reduction to enhance cell type/ state identification