Spectral Analysis of Mitochondrial Dynamics: A Graph-Theoretic Approach to Understanding Subcellular Pathology


Perturbations of organellar structures within a cell are useful indicators of the cell’s response to viral or bacterial invaders. Of the various organelles, mitochondria are meaningful to model because they show distinct migration patterns in the presence of potentially fatal infections, such as tuberculosis. Properly modeling and assessing mitochondria could provide new information about infections that can be leveraged to develop tests, treatments, and vaccines. Traditionally, mitochondrial structures have been assessed via manual inspection of fluorescent microscopy imagery. However, manual microscopy studies are labor-intensive and fail to provide a high-throughput for screenings. Thus, demonstrating the need for techniques that are more automated and utilize quantitative metrics for analysis. Yet, modeling mitochondria is no trivial task; mitochondria are amorphous, spatially diffuse structures that render traditional shape-based, parametric modeling techniques ineffective. We address the modeling task by using OrNet (Organellar Networks), a Python framework that utilizes probabilistic, graph-theoretic techniques to cast mitochondrial dynamics in the mold of dynamic social networks. We propose quantitative temporal and spatial anomaly detection techniques that leverage the graph connectivity information of the social networks to reveal time points of anomalous behavior and spatial regions where organellar structures undergo significant morphological changes related to a relevant change in environment or stimulus. We demonstrate the advantages of these techniques with the results of exhaustive graph-theoretic analyses over time in three different mitochondrial conditions. This methodology provides the quantification, visualization, and analysis techniques necessary for rigorous spatiotemporal modeling of diffuse organelles.