Artificial neural network: a powerful tool in associating phenolic compounds with antioxidant activity of grape juices

Abstract

In vitro techniques are essential to assess the antioxidant potential of foods, although methods with different action mechanisms make troublesome data analysis. This article describes the use of artificial neural network (ANN) to associate phenolic compounds with antioxidant activity in vitro (AOX) of grape juices. A multilayer perceptron (MLP) ANN was obtained with 28 phenolics quantified, as input layers, and AOX measuring by DPPH, ABTS, FRAP, H2O2, and β-carotene/linoleic acid bleaching assay (βCLA) methods, as output layers. To improve discussion in food sciences, the ANN results were compared with Pearson’s correlation and principal component analysis (PCA), methods largely used in food studies. Pearson’s technique showed correlations between antioxidant methods and some of the phenolic compounds, but with limitations. PCA proved to be a more powerful method than Pearson’s correlation, as it positively associated 13 phenolics with four out of five antioxidant methods. The MLP-ANN allowed simultaneous association of 19 individual phenolics, while a single hidden layer predicted 15 phenolics with simultaneous action in all AOX methods. The power of association was: ANN > PCA > Pearson. It was evidenced that ANN is a powerful tool for screening antioxidants in different AOX systems, which is applicable in health interests. Graphical abstract: [Figure not available: see fulltext.]