abstract: Systems biology efforts require accurate, cell-type specific information about the shape and distributions of subcellular structures and the distributions of proteins and other macromolecules in order to be able to capture and simulate cellular spatiotemporal dynamics.
We have developed tools to build generative models of cell organization directly from microscope images of many cells. Our open source system, CellOrganizer (http:/CellOrganizer.org), contains components that can build probabilistic generative models of cell, nucleus, organelle and protein distributions. The models also capture heterogeneity within cell populations. A critical challenge in constructing these models is to be able to learn the dependence of these distributions upon each other, i.e., the spatial relationships between different components. As these relationships are learned, generative models can be created from images of different proteins and organelles and then combined to create synthetic cells having many more components than can be imaged together.
An important use for these models is to generate synthetic cell shapes and organelle distributions that can be used as geometries for cell simulations. This permits a structured exploration of the dependencies of cellular biochemistry upon cell morphology and organization.