Date
Dec 2, 2024, 3:00 pm4:00 pm

Details

Event Description

The avalanche of experimental technologies for multi-modal profiling of tissues at single cell resolution is motivating the development of a new generation of mathematical representations for biology. Following the success of transformers and large language models in natural language processing but also protein structure predictions, it is being proposed that black box deep network strategies may evolve toward foundation models that will facilitate systematic “computation” of cells’ and tissues’ functions. But the new rich data also allow many biological phenomena, including cell state space, cellular differentiation, gene regulation, epigenetics or morphogenesis, to be represented in a principled and quantitative fashion. In this talk we will introduce notions of quantitative cellular manifolds (or atlases), and develop inference of temporal dynamics over them. Based on such manifolds we will introduce models for genome regulation underlying cellular differentiation dynamics. We will showcase the use of quantitative manifold models in redefining the quantitative regulatory impact of DNA methylation in mammalian development, in tracking human hematopoietic differentiation in absolute time and in devising new diagnostic tools for precise blood-test-based differential diagnosis of cytopenic patients. We will argue that at the basis of an effective foundation model for biology, whether implemented as a deep network or not, we need quantitative representations of the major biological constituents. Thanks to modern experimental technologies such representations are now within reach.

Event Category
QCB Seminar Series