Talk by Elena Sabbioni at BioInference 2023.
[ Ссылка ]
Abstract: RNA velocity is a biological quantity that, exploiting the gene expression of cells at different stages of maturity, provides insights into how cells differentiate along the development path. Gene expression is controlled by transcription, splicing, and degradation, which can be modeled by a stochastic chemical reaction network. Experimental data from single-cell RNA sequencing is generally used to estimate the rate constants of the network and consequently derive RNA velocity. The most known method to infer RNA velocity’s parameters is scVelo (Bergen V. et al, “Generalizing RNA velocity to transient cell states through dynamical modeling”, Nature biotechnology, 2020). This technique has been already criticized from many prospectives, and we also show that the current model formalization suffers from identifiability problems and it is unable to accurately recover the underlying transcription dynamics from simulated data. In contrast, we propose a simplified and mathematically better founded model, based on Bayesian statistic, that ensures parameter identifiability without relying on artificial preprocessing steps. Our approach relies on a data distribution motivated by reaction network theory and assumes that cells of the same type follow common dynamics.
![](https://s2.save4k.ru/pic/HsUP0Do3COQ/maxresdefault.jpg)