Reconstructing Waddington's landscape from data

Area of Science:

• Developmental Biology
• Computational Biology
• Systems Biology

Background:

• Zygote to organism development requires diverse cell types from a single progenitor.
• Existing developmental signaling network models are complex and difficult to fit to data.
• Landscape models offer an alternative by visualizing cell fate decisions as flows in an abstract topography.

Purpose of the Study:

• To develop a computational geometry framework for fitting dynamical landscapes directly to high-dimensional single-cell data.
• To map gene expression to developmental dynamics and characterize landscape features.
• To provide a general framework for understanding and controlling developmental dynamics.

Main Methods:

• Developed a computational geometry framework to fit dynamical landscapes to high-dimensional single-cell data.
• Modeled the time evolution of probability distributions in gene expression space.
• Applied the framework to multicolor flow-cytometry and RNA-seq data, including a stem cell system for neural tube patterning.

Main Results:

• Successfully constructed landscapes directly from high-dimensional single-cell data with minimal free parameters.
• Characterized dynamical features like fixed points, unstable manifolds, and basins of attraction.
• Recovered morphogen-dependent landscapes for neural progenitor types, showing signaling dependence and irreversible behavior after transient morphogen exposure.

Conclusions:

• The developed framework combines landscape model interpretability with direct data connection.
• It offers a powerful tool for understanding and controlling complex developmental dynamics.
• The method provides a general approach applicable to various single-cell datasets and biological systems.