Physics-informed deep learning characterizes morphodynamics of Asian soybean rust disease.

Area of Science:

• * Computational biology
• * Biophysics
• * Plant pathology

Background:

• * Traditional phenotypic screens for drug and biocide discovery rely on static, human-defined organism features.
• * Current methods lack the ability to dynamically characterize complex cellular responses to perturbations.
• * Understanding pathogen development is crucial for effective crop protection strategies.

Purpose of the Study:

• * To develop a novel computational framework for quantifying cell morphodynamics in response to chemical treatments.
• * To apply this framework to study the developmental perturbations of *Phakopsora pachyrhizi*, the Asian soybean rust pathogen.
• * To integrate unsupervised learning with biophysical modeling for a deeper understanding of biological systems.

Main Methods:

• * Development of a framework to analyze cell shape changes (morphodynamics) directly from images.
• * Description of population development in a 2D shape space (morphospace) using Fokker-Planck and tip growth models.
• * Application of condition-dependent parameters to model Waddington-type landscapes and phenotype transitions.

Main Results:

• * Characterization of diverse morphogenetic landscapes governing cell development under different conditions.
• * Identification of specific perturbations in the tip growth machinery responsible for observed phenotypic variations.
• * Demonstration of a novel integration of unsupervised machine learning and biophysical modeling techniques.

Conclusions:

• * The developed framework enables quantitative analysis of cell morphodynamics, offering insights beyond static phenotypic features.
• * This approach provides a powerful tool for interpreting drug-induced developmental changes in pathogens like *Phakopsora pachyrhizi*.
• * The integration of computational and biophysical methods presents a widely applicable strategy for biological discovery.