Interpreting biosensor data via backpropagation

Abstract

The authors present a discussion of the use of backpropagation to deconvolute fluorescence spectra obtained from solutions of amino acids in order to estimate concentration. Specifically, mixtures of tryptophan and tyrosine are considered. The input to the net is the fluorescence intensity at a number of wavelengths. The output from the net is the concentration of the fluorophores present. The fluorescence spectra considered are nonlinear, and it is shown that backpropagation is superior to standard linear techniques in estimating the solution concentrations. Although a specific sensor is treated, the approach taken is general and can be applied to other sensors used in the process industries.