FLUORESCENCE SPECTROSCOPY ANALYSIS THROUGH CONVOLUTIONAL NEURAL NETWORKS

Abstract

An important analysis in fluorescence spectroscopy is molecular dynamics, in particular, quantification of diffusion coefficient. In this research, we apply a convolutional neural network (CNN) to determine the diffusion coefficient of freely diffusing molecules. We construct the CNN based on our concurrent study of model identification of autoregression moving average (ARMA) time series through CNN, which illustrates the application of CNN to temporal data. We train the network with purely simulated data under a quadratic loss function. The performance of our network is validated using simulated data and data from lipid bilayer experiments. For simulated validation data, we find that our network performs better at 2,500 frames against the conventional non-linear least squares curve fitting approach in Imaging FCS method, in terms of mean squared error. We demonstrate using scatter plots that our network predictions for lipid bilayer data are comparable to Imaging FCS.