Lineage-aware stochastic modeling reveals gene-expression dynamics in development and disease

Area of Science:

• Computational Biology
• Genomics
• Developmental Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) data is often treated as static snapshots, neglecting dynamic gene expression changes along cell lineages.
• Existing methods struggle with the sparse and overdispersed nature of scRNA-seq counts and often use imprecise Gaussian assumptions.
• Reconstructing cell lineage phylogenies offers a framework to study transcriptional changes during development, differentiation, and disease.

Purpose of the Study:

• To develop a probabilistic framework, LaVOUS (Lineage-aware Variational Ornstein-Uhlenbeck Single-cell RNA-seq analysis), for analyzing gene expression dynamics in the context of cell lineage trees.
• To enable likelihood-based testing of cellular heritability and branch-specific gene expression shifts.
• To reconstruct latent expression histories phylogenetically.

Main Methods:

• LaVOUS couples lineage-based models of latent dynamics (Brownian motion, Ornstein-Uhlenbeck processes) with negative-binomial observation models.
• It employs scalable variational inference for efficient analysis of sparse count data.
• The framework integrates cell lineage tracing with single-cell transcriptomic data.

Main Results:

• In simulations, LaVOUS demonstrated superior performance over Gaussian methods in detecting lineage-associated expression changes and reconstructing expression histories.
• LaVOUS successfully identified lineage-associated gene expression changes in metastatic lung cancer, class-switching B cells, and developing brain tissues.
• Specific applications revealed insights into metastatic progression, B-cell isotype switching, and neuronal differentiation.

Conclusions:

• LaVOUS provides an expressive and scalable framework for modeling sparse count data on lineage trees, advancing the study of single-cell expression dynamics.
• It establishes a foundation for investigating gene expression changes in developmental and disease contexts.
• Future extensions can incorporate multi-gene regulation, lineage uncertainty, and multi-modal data integration.