Evolving dynamical landscapes of cell fate transitions

Epigenetic landscape models describe cell fate transitions in a low-dimensional state space, where attractors represent cell types, and stochastic jumps and bifurcations drive cellular decisions. While such an approach has produced quantitative and predictive descriptions (Corson and Siggia, 2012; Camacho-Aguilar et al; 2021, Sáez et al, 2022), current applications are designed in a largely ad hoc manner. We thus propose a method for systematic generation of landscape models for various systems. For constructing the landscape, we combine gradient and rotational vector fields, composed of locally weighted flow elements. This results in an expressive yet interpretable model. In contrast to existing polynomial landscapes, the components of our model are at the scale of individual cell types, allowing flexible topology optimization with minimal assumptions on landscape geometry and differentiation routes. To optimize landscapes based on data, we use an evolutionary algorithm, generating an ensemble of solutions that reveals both known and novel landscapes in terms of topology and bifurcations and provides statistics for parameters. I will present the application of our method to two systems: neuromesoderm differentiation and the segmentation clock.