Using a Bayesian framework for inverse problems where the target is continuous parameter functions, I present a method for inference and uncertainty quantification on the physical properties that affect macromolecular bond motion. A model of bond motion as overdamped Brownian motion suggests position-dependent diffusivity and bond forces are targets for reconstruction. I will describe how one may use an empirical Bayesian method for selection of regularization parameters and for approximating the reconstruction error.

I will also give an overview of a class of (true) Bayesian methods for model selection based on the minimization of prediction error. As an application of these methods I will talk about the selection of memory for multi-step Markov chains.