Point process models for localization and interdependence of punctate cellular structures

Abstract

Background Accurate representations of cellular organization for multiple eukaryotic cell types are required for creating predictive models of dynamic cellular function. To this end, we have previously developed the CellOrganizer platform, an open source system for generative modeling of cellular components from microscopy images. CellOrganizer models capture the inherent heterogeneity in the spatial distribution, size, and quantity of different components among a cell population. Furthermore, CellOrganizer can generate quantitatively realistic synthetic images that reflect the underlying cell population. A current focus of the project is to model the complex, interdependent nature of organelle localization. Results We built upon previous work on developing multiple non-parametric models of organelles or structures that show punctate patterns. The previous models described the relationships between the subcellular localization of puncta and the positions of cell and nuclear membranes and microtubules. We extend these models to consider the relationship to the endoplasmic reticulum, and to consider the relationship between the positions of different puncta of the same type. Our results do not suggest that the punctate patterns we examined are dependent on ER position or inter- and intra-class proximity. With these results, we built classifiers to update previous assignments of proteins to one of 11 patterns in three distinct cell lines. Conclusion Our generative models demonstrate the ability to construct statistically accurate representations of puncta localization from simple cellular markers in distinct cell types, capturing the complex phenomena of cellular structure interaction with little human input. This protocol represents a novel approach to vesicular protein annotation, a field that is often neglected in high-throughput microscopy. These results suggest that spatial point process models provide useful insight with respect to the spatial dependence between cellular structures.

Learning unsupervised feature representations for single cell microscopy images with paired cell inpainting

“…To further test the possibility that the unsupervised analysis in the feature space reveals higher resolution functional information than expert annotation, we asked if our features could be used to discover higher-resolution subclasses of proteins that are difficult for experts to annotate visually. First, previous work with vesicle-localized proteins in the Human Protein Atlas has suggested that these proteins can be distinguished into many subclasses (2,51). We clustered features for human proteins annotated to localize to the vesicle only, and found structure in the feature representation that corresponds to visually-distinct vesicle patterns (Abundantly and more evenly distributed in the cytosol, concentrated closer to the nucleus or sparse puncta, Supplementary Figure 2, Clusters A, B and C respectively).…”

Imaging and spatial analysis of hematopoietic stem cell niches

“…As seen by the number of emerging studies, there is an obvious need for quantitative methods that can capture and describe complex spatial associations . The consideration of cells in tissues as point patterns enables the use of point processes statistical theory, which provides validated mathematical methods widely employed in other fields dealing with spatially organized systems .…”

Robust classification and detection with applications in biomedical images

“…The reason for such a difference is we focus on binary classification other than a generalized multi-categorical classification problem in this section. Table 2.1, the expected cost EC(t 1 ,t 2 ) can be formulated with thresholds t 1 and t 2 [23]: 16) where FNR , T NR, T PR, FPR are the false negative rate, the true negative rate, the true positive rate and the false positive rate while RP and RN are the rejection rates on the positive and the negative samples, respectively. p(P) and p(N) are the prior probabilities of the positive and negative classes, respectively.…”

“…Different pairs of (P, R) determine different LBP configurations. The most typical settings are (8,1), (12,2) and (16,4) [48].…”

“…In the last decade, machine learning approaches have achieved great success towards applications in biomedical images (see e.g., [12] for a general review). For instance, the structures and functions of different proteins were studied based on their locations in specific organelles using image classification models [13][14][15][16]. In [17,18], mixture patterns of protein distributions were recognized and [19] proposed to identify protein distribution patterns in time series images.…”