Single molecule techniques continue to open new windows to the inner workings of molecular biology, by allowing us to follow individual proteins at work both in vitro and in vivo.
However, instrumental noise, cross-sample variations, and the inherent randomness of Brownian motion and low copy number chemistry often makes for noisy data that can be challenging to interpret. A very common problem is to analyze noisy time series with abrupt changes, reflecting for example binding events or conformational changes in a protein complex under study.
In this talk, I will describe my ongoing attempts to tackle such problems with a combination of physical modeling and a machine learning technique known as ensemble learning or variational Bayes. In particular, I will discuss an application to single particle tracking data from the lab of Johan Elf at Uppsala University, where we extract in vivo binding kinetics from changes in the diffusion constant of small cytoplasmic proteins in E coli.
