CellOrganizer: Image-Derived Models of Subcellular Organization and Protein Distribution

Abstract

This chapter describes approaches for learning models of subcellular organization from images. The primary utility of these models is expected to be from incorporation into complex simulations of cell behaviors. Most current cell simulations do not consider spatial organization of proteins at all, or treat each organelle type as a single, idealized compartment. The ability to build generative models for all proteins in a proteome and use them for spatially accurate simulations is expected to improve the accuracy of models of cell behaviors. A second use, of potentially equal importance, is expected to be in testing and comparing software for analyzing cell images. The complexity and sophistication of algorithms used in cell-image-based screens and assays (variously referred to as high-content screening, high-content analysis, or high-throughput microscopy) is continuously increasing, and generative models can be used to produce images for testing these algorithms in which the expected answer is known.

Introduction

As traditional reductionist paradigms of biomedical research increasingly give way to systems approaches, the need to build predictive models that synthesize large amounts of information from potentially diverse sources is becoming critical. Most such current models take the form of transcriptional regulatory networks, protein–protein interaction maps, or biochemical reaction simulations. These typically do not consider spatial organization of cells or tissues. Important advances came with systems such as MCell (Stiles et al., 1998), which allowed models to be constructed using mesh representations of cells built from electron microscope images, and the Virtual Cell (Loew and Schaff, 2001), which allowed appropriately processed images to provide surface area and volume for its compartmental models. Ontologies such as the genome ontology (GO) can be used to describe protein attributes, including location, primarily at a major organelle level. Such assignments can also be used to create compartmental models (e.g., http://biologicalnetworks.net/tutorials). However, compartmental models suffer from some important limitations, in that they treat all molecules within each compartment as being homogenously distributed, and they do not allow appearance, disappearance, fission or fusion of compartments.

Given the energy expended by cells to maintain their subcellular organization, and the many defects that are associated with alterations in it, models that do not accurately reflect subcellular organization are unlikely to perform satisfactorily at predicting complex cell behaviors or how they respond to changes in conditions. There is therefore a need for computational models that accurately represent the number, size, shape, and positions of subcellular structures, the spatial relationships between different structures, and how proteins (and other molecules) are distributed between them (Murphy, 2010, Murphy, 2011). In addition, there is a need for a mechanism for representing how all of these vary within a population of cells of a single cell type, within a single cell type under different conditions, among different cell types, and among different organisms. Such models can not only capture cell behavior but can also be an important step in understanding that behavior, since, for example, a sufficiently detailed model helps distinguish aspects that are conserved and presumably necessary from those that are highly variable and potentially not necessary.

In considering how to build such models, we can distinguish descriptive models, which allow one to recognize what state a particular cell is in, from generative models, which can also synthesize new examples of cells in particular states. We can also distinguish theoretical or conceptual models, which posit a particular structure based on a generalized understanding, from data-driven models that are learned from data and capture both general behavior and variation in that behavior.

My focus in this chapter will be primarily on methods developed in my group that have been used to learn generative models of cell organization and protein distribution from two-dimensional and three-dimensional fluorescence microscope images (Zhao and Murphy, 2007, Rohde et al., 2008a, Rohde et al., 2008b, Peng et al., 2009, Shariff et al., 2010a, Shariff et al., 2011, Peng and Murphy, 2011). We have recently grouped these methods as part of the open source CellOrganizer project (http://cellorganizer.org), which includes collaborations with a number of investigators studying particular cell systems.