I³MS - Theis Seminar
Prof. Dr. Dr. Fabian Theis - Modeling cell-to-cell heterogeneities
Institute of Computational Biology, Technical University of Munich
Cell-to-cell variations in gene expression underlie many biological processes. With single-cell expression techniques finally becoming available in order to robustly observe these variations in an unbiased manner, new, biologically relevant subpopulations may be observed. However, without multi-cell averaging as in population data, previously ignored noise sources may confound the actual biological signal. The observed variability in single-cell data is generated by a multitude of latent processes such as technical noise, censoring effects and biological confounders (e.g. cell cycle effects). Here we show how to deconvolve the observed heterogeneity by explicitly modeling these latent processes, and illustrate illustrate this on single-cell qPCR and RNAseq data from cellular differentiation decisions. In the second part of the talk, I will then illustrate how to include dynamic information for smaller single-cell data sets by combining mixture models and ordinary differentials equations. We use this ODE-constrained mixture modeling to study NGF-induced Erk1/2 phosphorylation in primary sensory neurons and reconstruct static and dynamical subpopulation characteristics across experimental conditions.