Application of artificial neural networks to the prediction of the antioxidant activity of essential oils in two experimental in vitro models

Introduction tioxidant properties of essential oils (EOs) have been on the centre of intensive research for their potential use as preservatives or nutraceuticals by the food industry. The enormous amount of information already generated on this subject provides a rich field for data-miners as it is conceivable, with suitable computational techniques, to predict the antioxidant capacity of any essential oil just by knowing its particular chemical composition. To accomplish this task we here report on the design, training and validation of an Artificial Neural Network (ANN) able to predict the antioxidant activity of EOs of known chemical composition. s ilayer ANN with 30 input neurons, 42 in hidden layers (20 → 15 → 7) and one neuron in the output layer was developed and run using Fast Artificial Neural Network software. The chemical composition of 32 EOs and their antioxidant activity in the DPPH and linoleic acid models were extracted from the scientific literature and used as input values. From the initial set of around 80 compounds present in these EOs, only 30 compounds with relevant antioxidant capacity were selected to avoid excessive complexity of the neural network and minimise the associated structural problems. s and discussion N could predict the antioxidant capacities of essential oils of known chemical composition in both the DPPH and linoleic acid assays with an average error of only 3.16% and 1.46%, respectively. We also discuss on the contribution of different compounds to these chemical activities. sions results confirm that artificial neural networks are reliable, fast and cheap tools for predicting antioxidant activity of essential oils from some of its components and can be used to model biochemical properties of complex natural products including the prediction of parameters associated with nutraceutical properties of food ingredients.