Analysis of molecular diffusion by machine learning-aided FCS

Abstract

It is important to understand how bio-molecules diffuse and are transported inside the cells in order to reveal the role of these molecules. Fluorescence correlation spectroscopy (FCS) is one of the methods to analyze such diffusion phenomena. In FCS, the diffusion coefficient of the target molecule can be estimated from the time-series data of fluorescence intensity by fitting the auto-correlation function with a model equation. Since FCS requires long time-series data to estimate the diffusion coefficient, we suggest a new FCS method, machine learning-aided FCS. In this new method, we train the neural network to understand the relationship between the time-series data of the intensity and the diffusion coefficient. Our approach can estimate the diffusion coefficient with time-series data that is 1/5 of the conventional FCS.