2025 Theses Doctoral

Probabilistic Modeling of Alternative Splicing in Single Cells

Isaev, Keren

Alternative splicing (AS) is a fundamental mechanism by which single genes produce multiple RNA and protein isoforms, expanding the functional repertoire of the genome. While AS plays established roles in regulating cell identity and key biological processes, its contribution to cellular heterogeneity at single-cell resolution remains poorly characterized.

Although single-cell RNA sequencing (scRNA-seq) has transformed our ability to profile gene expression in individual cells, the analysis of splicing variation has lagged behind. Sparse read coverage and reliance on predefined expression patterns or annotated cell types hinder the detection of splicing-driven cell states, potentially obscuring important axes of biological variation.

This dissertation addresses these challenges through three probabilistic frameworks tailored to single-cell splicing analysis. Leaflet, a Binomial mixture model, infers discrete splicing states directly from junction usage without requiring cell-type annotations. LeafletFA extends this approach to atlas-scale datasets with a Beta-Dirichlet factor model, revealing latent splicing programs and conserved age associated signatures across mouse and human tissues. SpliceVI integrates splicing with gene expression in a unified deep generative model, employing a partial encoder and Beta-Binomial likelihood to accommodate sparse splicing data while enabling accurate cross-modal imputation and refined delineation of cellular subtypes.

Together, these methods position AS as a critical yet underexplored contributor to cellular identity and provide the computational foundation for systematically characterizing splicing variation in single-cell transcriptomes.