QLS Seminar Series - Paul Francois

Waddington Landscapes in the Age of Machine Learning

Paul Francois, Universite de Montreal
Tuesday December 2, 12-1pm
Zoom Link: https://mcgill.zoom.us/j/87078928687
In Person: 550 Sherbrooke, Room 189

Abstract: Waddington’s epigenetic landscape remains a central concept in developmental biology, illustrating how cells follow robust, canalized paths toward distinct fates. Despite the rise of high-resolution techniques like single-cell RNA sequencing, translating these data into actual landscapes remains a major challenge. In this talk, I will introduce EvoScape, a biologically motivated framework that constructs pseudo-landscapes reflecting key properties of differentiation. EvoScape blends concepts from biophysics and machine learning to produce intuitive, mechanistic models of developmental trajectories. I will illustrate how EvoScape can generate abstract models of vertebrate segmentation, and—more concretely—how it reconstructs a landscape of neuromesodermal differentiation directly from single-cell data, revealing a structure consistent with known developmental dynamics. Finally, I will explore how learning dynamics in Generalised Hopfield Networks themselves exhibit Waddington-like behaviors, pointing to intriguing parallels between biological and computational learning.