Tools4Cells: Machine-learning aided morphodynamics characterization of stem cell differentiation using label-free microscopies

University Of Pittsburgh, Pittsburgh PA

Investigators

Abstract

It is a great wonder of Nature that an embryo can develop into an organism composed with different types of cells, and different types of cells can interconvert under certain conditions. Mechanistic understanding and manipulation of conversion between different cell types emerge as an exciting frontier of research. This area of research has the potential to contribute to our understanding of development as well as to innovations in tissue and organ regeneration. The goal of this project is to develop computational tools for analyzing movies of cells to learn how individual cells change from one type to another one. It is analogous to tracing the routes people take for driving from one city to another one, so one can identify the dominant routes and try to direct traffic to specific route. The Broader Impacts of the work include the intrinsic merit of the research as the platform could prove useful to a number of other projects. Research opportunities will be made available to undergraduate and graduate students along with post-doctoral researchers. Outreach effort will involve the use of YouTube to promote science to the general public. Cells of different phenotypes are characterized by distinct morphology and gene expression patterns. When subject to specific stimuli and microenvironments they can transit between distinct phenotypes, reprogramming to become induced pluripotent stem cells. Mechanistic understanding of such cell phenotypic transition (CPT) processes, however, suffers from the challenge that snapshot data cannot provide temporal information. Live-cell imaging provides an alternative approach for monitoring transitions of individual cells continuously over time. The project aims to develop a generally applicable platform and associated computer package that (1) allows typical cell biology labs to apply the platform and study cellular processes without requirements of a strong background in machine-learning, image analyses, or dynamical systems theories; (2) extends to various imaging platforms and modalities, especially quantitative phase imaging-based label-free imaging that has gained popularity together with machine-learning approaches. Beyond the traditional bioimage informatics approaches, the novelty here is for mechanistic studies of CPT dynamics in the framework of dynamical systems theory/systems biology. In addition, inspired by the critical role the image database ImageNet has played on the recent advance of artificial intelligence development, the researchers will provide annotated cell images generated from this project as benchmarks for the community to develop and test image analysis approaches. The developed framework will be generally applicable to study a large number of cell phenotype conversion processes. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.