Artificial neural networks for quantification in unresolved capillary electrophoresis peaks

Artificial neural networks (ANN) have been applied to the resolution of ove rlapping capillary electrophoresis peaks of amino acid derivatives labelled with 1,2-naphthoquinone-4-sulfonate (NOS). The separation was performed wi th a fused-silica capillary and the corresponding 3D-electropherograms were recorded in a range from 225 to 550 nm with a diode array detector (DAD). Since complete resolution of all the analytes was not accomplished, a chemo metric approach was used to improve the quantification mathematically. In t he present case, a three-layer back propagation (BP) ANN with a sigmoid tra nsfer function was built in order to perform the amino acid determination. The inputs of the ANN were the spectra or the electropherograms of each sam ple and the outputs were the concentrations of the amino acid derivatives i n the overlapping peaks to be predicted. The results were compared with tho se from partial least squares regression (PLS) and, in general, ANN provide d better predictions than PLS.