Image‐derived, three‐dimensional generative models of cellular organization

Abstract

Given the importance of subcellular location to protein function, computational simulations of cell behaviors will ultimately require the ability to model the distributions of proteins within organelles and other structures. Towards this end, statistical learning methods have previously been used to build models of sets of two-dimensional microscope images, where each set contains multiple images for a single subcellular location pattern. The model learned from each set of images not only represents the pattern but also captures the variation in that pattern from cell to cell. The models consist of sub-models for nuclear shape, cell shape, organelle size and shape and organelle distribution relative to nuclear and cell boundaries, and allow synthesis of images with the expectation that they are drawn from the same underlying statistical distribution as the images used to train them. Here we extend this generative models approach to three dimensions using a similar framework, permitting protein subcellular locations to be described more accurately. Models of different patterns can be combined to yield synthetic multi-channel image containing as many proteins as desired, something that is difficult to obtain by direct microscope imaging for more than a few proteins. In addition, the model parameters represent a more compact and interpretable way of communicating subcellular patterns than descriptive image features, and may be particularly effective for automated identification of changes in subcellular organization caused by perturbagens.

Learning and visualizing statistical relationships between protein distributions from microscopy images

“…We utilized 233 2D images, selected at random and segmented using the software described in [13]. Finally, the third dataset used was a set of HeLa cells obtained using fluorescence microscopy images [14]. In total a set of 70 images were segmented semi automatically using a simple thresholding approach.…”

Building cell models and simulations from microscope images

“…For nuclei, parametric approaches have the advantage discussed above that they are compact and easy to learn. Examples include 2D medial axis models [14] and 3D cylindrical surface models [18]. Models of nuclear shape variation are created using these approaches by estimating the parameters for each nucleus, and then constructing a statistical model of the distribution of those parameters over all nuclei available for training.…”

“…The first dependent model of cell and nuclear shape was created for 2D images using a simple ratiometric approach [14] in which the ratio of the distance from the center of the nucleus to the cell boundary and to the nuclear boundary was calculated as a function of the angle from the major axis of the cell. This was later extended to 3D [18]. As with the parametric nuclear models, these ratiometric models are not suitable for complex cell shapes.…”

“…In this case, the task, given an image, becomes to identify the distribution of Gaussian objects most likely to have given rise to that image. This standard Gaussian mixture modeling approach has been used to identify positions and sizes for vesicular organelles, and the results used to produce statistical models of the number of organelles per cell and their variation in size [14, 18]. An example of an image synthesized from such a model is shown in Figure 5.…”

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

“…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 (5–11). 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.…”