engleski

5'- ribonucleotides mixtures resolutions by ANNs using UV spectroscopic data

Artificial neural networks such as multilayer perceptrons or generalized feedforward networks can detect characteristic patterns and generalize hidden relationships in chemometric problems like 5’-ribonucleotides mixtures resolution using UV spectroscopic data. Resolving of the spectral overlapping was successfully done with simulated spectra, as well as with standard solution mixtures of 5’-CMP, 5’-GMP, 5’-UMP and 5’-AMP nucleotides. Standard solutions of 5’-ribonucleotides (Sigma Co.) were prepared in a phosphate buffer solution pH 7.0. Absorbances of all samples were acquired from 210 to 300 nm in 1.0 nm intervals at a constant scanning rate of 180 nm/min. Artificial neural networks have proved to be successful in the various problems connected with, for instance a deconvolution of pyrolysis mass spectra, or a quantitative discrimination of the pure ribonucleotides spectrophotometric data. On the other hand, the 5’-ribonucleotides are high added-value products used as flavor enhancers (5’-GMP, 5’-IMP), and have important applications in pharmaceutical industry. In this work, simulated spectroscopic data and experimentally obtained 5’-ribonucleotides spectra of the pure components and of their mixtures were used for the quantification of the mixtures ingredients. Those complex mixtures could not be expressed simply in terms of additive combinations of the spectra describing the pure components of the mixtures and their relative concentrations. So, in simulations a complex nonlinear superposition were applied that could be described by the following equation: z=sin(y1)+sin(y2). For training, cross validation and testing of ANNs, each input (i.e. normalized z spectrum) was paired with the desired outputs (normalized y1 and y2 spectra, respectively). Alternatively, the sinus of a numerical summation of measured pure components spectra, or experimentally recorded mixtures spectra were also used for the evaluation of the basic idea. The computer simulation results had showed a quite impressive ANN power of the binary mixtures resolution and furthermore, the very promising results were obtained in the estimates of the concentrations of the components in the experimentally prepared binary, ternary and quaternary 5’-ribonucleotides mixtures, based on an ANN model of UV spectroscopic data.

artificial neural networks; 5’-ribonucleotides; spectroscopy

nije evidentirano