RNA velocity is a biological quantity involved in the interpretation of cellular differentiation, exploiting the evolution of gene expression of cells at different stages of maturity during the evolution path. Gene expression is regulated through three fundamental processes, transcription, splicing and degradation, which are modelled as a stochastic chemical reaction network. The development process among the cells is triggered by the switch of some genes, which is modelled through a binary modification of the gene-specific transcription rates. The available data to estimate the rate constant of the network, are collected by single-cell RNA sequencing. The principal method that is used in RNA velocity for the estimation procedure is scVelo (Bergen V. et al, “Generalizing RNA velocity to transient cell states through dynamical modelling”, Nature biotechnology, 2020). This approach has already been criticized over various aspects and we also show that, starting from simulated data, it is not able to recover the underlying parameters that describe the transcription dynamic. We propose a model that has some modifications and is also mathematically better founded and that, differently from scVelo, ensures the identifiability of the parameters. Among the fundamental differences with scVelo, we have the substantial absence of artificial preprocessing steps, we make use of a data distribution that is motivated by reaction network theory and assume that the cells of the same type follow common dynamics.