Chemoinformatic modelling of the antioxidant activity of phenolic compounds

Abstract

BACKGROUND: Antioxidants are chemicals used to protect foods from deterioration by neutralizing free radicals and inhibiting the oxidative process. One approach to investigate the antioxidant activity is to develop quantitative structure–activity relationships (QSARs). RESULTS: A curated database of 165 structurally heterogeneous phenolic compounds with the Trolox equivalent antioxidant capacity (TEAC) was developed. Molecular geometries were optimized by means of the GFN2-xTB semiempirical method and diverse molecular descriptors were obtained afterwards. For model development, V-WSP unsupervised variable reduction was used before performing the genetic algorithms–variable subset selection (GAs-VSS) to construct the best five-descriptor multiple linear regression model. The coefficient of determination and the root mean square error were used to measure the performance in calibration (R2 = 0.789 and RMSEC = 0.381), and test set prediction (Q2 = 0.748 and RMSEP = 0.416), along several cross-validation criteria. To thoroughly understand the TEAC prediction, a fully explained mechanism of action of the descriptors is provided. In addition, the applicability domain of the model defined a theoretical chemical space for reliable predictions of new phenolic compounds. CONCLUSION: This in silico model conforms to the five principles stated by the Organisation for Economic Co-operation and Development. The model might be useful for virtual screening of the antioxidant chemical space and for identifying the most potent molecules related to an experimental measurement of TEAC activity. In addition, the model could assist chemists working on computer-aided drug design for the synthesis of new targets with improved activity and potential uses in food science.