Bayesian Nonparametrics for Fluorescence Methods

Abstract

Fluorescence methods are widely used across Biophysics. Fluorescence signals contain important information on biomolecular chemistry and dynamics. This information can be further applied to characterize and monitor complex biological processes. However, it is normally difficult to establish accurate parametric models for biophysical systems on account of the noise inherent to single molecule methods. For this reason, we introduce Bayesian nonparametrics (BNPs) that go beyond parametric modeling where models of single molecule dynamics are assumed from the onset. BNP approaches allow us to introduce an explicit mathematical formulation that quantitatively describes the functional relation from the observation data. This method provides a purely empirical identification of sample fluorescence properties that can serve as input in future coarse-grained modeling simulations. BNPs are poised to play an important role in the analysis of fluorescence signals since so few model features-such as the number of states of a single molecule in any time trace-are known a priori. Moreover, BNPs provide full posteriors over models and parameters in comparison to existing Bayesian approaches that provide only distributions over parameters.