Integration and harmonization of cell shape images for generative modeling

Abstract

As cell imaging grows in scale, precision, and complexity, data integration and harmonization become increasingly important for studying cell-material interactions. Quantitative understanding of how cells respond to mechanical cues, such as substrate stiffness and topography, is often limited by differences in experimental conditions and imaging formats. This study presents a framework that combines compact, interpretable cell shape models with generative artificial intelligence to harmonize 2D and 3D immunofluorescent datasets within defined experimental contexts. By efficiently capturing morphology and associated biological features, the approach enables generation of realistic synthetic cells, including rare or intermediate phenotypes, to augment machine-learning analyses and support scalable in silico studies. This work advances data-driven investigation of cellular responses to biomaterial-derived mechanical cues.