Antioxidant capacity prediction: Combining individual compound capacities to predict plant-extract capacities

The antioxidant activity of plant extracts offer potential for preventing and managing degenerative diseases linked to oxidative stress. Whilst the structure-function relationship of individual compound antioxidant capacities is well-established, accurate prediction of the overall antioxidant capacity of complex mixtures such as plant extracts, remains challenging. In this study, we sourced a data set of 68 plant extracts with empirically determined antioxidant capacities via the 2,2-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid) diammonium salt assay (ABTS assay) and paired quantitatively determined metabolite profiles. Using a previously developed Trolox equivalent antioxidant capacity (TEAC) model, we predicted the antioxidant capacities of each phytochemical within these profiles. We then employed polynomial regression with k-fold cross-validation (k = 10) to develop a model predicting the antioxidant capacity of the plant extracts. The model, which utilised the count and sum of individual compound capacities of antioxidant-capable phytochemicals to predict log 10 (TEAC), achieved an R 2 of 92.28 % and a 10-fold cross-validated R 2 of 74.49 %. When transformed back to TEAC (mM), the model resulted in an R 2 of 94.59 %, with 77.9 % of predictions within 20 % of their true values. These results demonstrate the utility of statistical models in predicting individual phytochemical antioxidant capacities and their contributions to food functional properties. Our model represents a significant advancement in predicting plant-extract antioxidant capacities from their phytochemical compositions, with implications for optimising functional food value through targeted modulation of phytochemical profiles or strategic blending (fortification) of plant extracts. • Two models were developed for prediction of plant-extract antioxidant capacities. • Prediction of TEAC from compound antioxidant capacity achieved 10-fold R 2 of 74.49 %. • This model provides a novel strategy for predicting mixture antioxidant capacity.