Bayesian analysis on mixture models, for understanding the process of myosin binding to the thin filament

Understanding how access is granted to myosin by the actin thin filament has not been fully understood yet. The process of thin filament activation is explored by developing a new variation of hidden Markov models to extract dynamic information from image data and to establish how many myosins are present in an activated region against time. Hidden Markov models supply an extension to mixture models in such a way that they allow for spatial data. The novelty lies in the model allowing for spatial information in the image to be encoded through contextual constraints of a neighbourhood structure based on three nearest neighbours. Furthermore, for the purpose of Bayesian inference about the unknown number of K components, the Metropolis-Hastings algorithm is employed. The Bayesian analysis shows that, when compared to reversible jump Markov chain Monte Carlo, our proposed model provides a better alternative for the finite mixture model at capturing the behaviour of myosin binding to the thin filament. The estimated mean intensity values of uorescence from both models are exemplified in separate kymographs, where the variation in light intensity gives us information about how the myosin binding phenomenon is clustered or varies over time.