| Microtubules are cylindrical protein polymers constituting one of the components of the cytoskeleton - cell skeleton. They participate in a number of cellular functions such as maintaining cell shape, cell division and transport of various molecules. Microtubules are highly dynamic polymers, constantly changing in length through addition and loss of polymer units, a process called dynamics instability. Abnormal regulation of these dynamic behaviors has been associated to neuro-degenerative diseases such as Alzheimer's and FTDP-17, as well as cancer. Due to their importance in maintaining proper cell function, researchers study microtubule dynamics under different conditions such as the various concentrations of regulatory proteins and different drug treatments.; Despite remarkable advances in live cell imaging technologies, in most cases, analysis remains a painstaking, tedious and largely manual task, limiting the quantity and quality of analyzable data. Additionally, a number of assumptions are required to facilitate manual data collection task. In this work, we describe (i) computer vision methods to detect and quantify changes in microtubule length and (ii) pattern analysis methods modeling distinct behavioral characteristics within each experimental condition and common patterns across different conditions. We show that through computational detection and tracking of microtubules, each video produces significantly more usable data with improved accuracy and objectivity, and without requiring user-imposed assumptions. Furthermore, we show that the proposed methods permit calculation of new dynamics features that are impractical to collect manually. We are able to track microtubules despite considerable challenges manifested in microtubule videos. In particular, a path optimization technique is considered for handling microtubule tips that constantly move in and out of the focal plane during image acquisition. With substantial increases in available dynamics data in the form of individual microtubule tracks, we introduce novel hidden Markov model-based analysis capabilities. Facilitated by probabilistic similarity measures between individual microtubules and experimental conditions, as well as between conditions, we use a model based clustering approach for extracting common dynamics patterns instigated by different regulatory agents. Our experimental results confirm the applicability of computational contributions in microtubule dynamics research. We illustrate the use of estimated models in extraction of spatial and temporal patterns which offer previously unattainable insights to cellular processes. |
