Previous Article in event
Previous Article in congress
Next Article in event
Next Article in congress
Machine-Learning models to predict the antioxidant capacity of food
*
1, 2
,
1
,
2
,
1
,
2
,
1, 3
,
2
,
1, 2
Published:
12 December 2016
by MDPI
in MOL2NET'16, Conference on Molecular, Biomed., Comput. & Network Science and Engineering, 2nd ed.
congress USEDAT-02: USA-Europe Data Analysis Training Program Workshop, Cambridge, UK-Bilbao, Spain-Miami, USA, 2016
Abstract: The growing increase in the amount and type of nutrients in food created the necessity for a more efficient use in dietetics and nutrition. Flavonoids are exogenous dietary antioxidants and contribute to the total antioxidant capacity of the food. The current work aims to obtain optimal models to predict the total antioxidant properties of food by the ORAC method. A dataset based on the Database for the Flavonoid Content of Selected Foods and the Database for the Isoflavone Content of Selected Foods, was created. Different algorithms of artificial intelligence were applied, in particular Machine-Learning methods. They were employed using a R language. The performed study allowed to show the effectiveness of the models using structural-topologic features of Topological Substructural Molecular Design (TOPSMODE) in the databases. The proposed models can be considered, without overfitting, effective in predicting new values of ORAC, excepting the MultiLayer Perceptron (MLP) algorithm. The optimal model was obtained by the Random Forest (RF) algorithm, which presented the best R2 of the series (R2 = 0.9571313 for the training series and R2= 0.9247337 for the external prediction series).
Keywords: Flavonoid, Total antioxidant capacity, Artificial intelligence, Machine-Learning methods, Random Forest algorithm.
Comments on this paper
Marcus Scotti
10 January 2017
Consensus model
Dear authors, first of all, I would like to congratulate for the nice work. The models show higher values of coefficient of determination and prediction. I have a little comment about the number of significant digits in the coefficient of determination (R2): I think that three could be enough. I have a question: Since the authors used several models, do you will in the near try to perform a consensus model in order to improve the power of prediction.
Thanks
Best,
Marcus
Thanks
Best,
Marcus
Estela Guardado Yordi
11 January 2017
Dear Marcus,
Thank you very much for your comments and suggestions. We are working to improve predictions, using other methods of attribute selection, because we are interested in confirming the dependence and importance (in this field of food sciences) of the attributes that encode TOPS-MODE topological information. We have also thought about focusing on the RF algorithm and improving prediction.
Thanks
Best,
Estela
Thank you very much for your comments and suggestions. We are working to improve predictions, using other methods of attribute selection, because we are interested in confirming the dependence and importance (in this field of food sciences) of the attributes that encode TOPS-MODE topological information. We have also thought about focusing on the RF algorithm and improving prediction.
Thanks
Best,
Estela
