Multivariate statistical approaches for wine classification based on low molecular weight phenolic compounds
Version of Record online: 10 APR 2012
© 2012 Australian Society of Viticulture and Oenology Inc.
Australian Journal of Grape and Wine Research
Volume 18, Issue 2, pages 138–146, June 2012
How to Cite
CABRITA, M.J., AIRES-DE-SOUSA, J., GOMES DA SILVA, M.D.R., REI, F. and COSTA FREITAS, A.M. (2012), Multivariate statistical approaches for wine classification based on low molecular weight phenolic compounds. Australian Journal of Grape and Wine Research, 18: 138–146. doi: 10.1111/j.1755-0238.2012.00182.x
- Issue online: 25 MAY 2012
- Version of Record online: 10 APR 2012
- Manuscript received: 17 July 2011; Revised manuscript received: 22 December 2011; Accepted: 22 January 2012
- artificial neural network;
- phenolic compound;
- principal component analysis;
- variance partitioning;
Background and Aims: Phenolic compounds influence colour, flavour and astringency of wines. Technology and grape variety are the main factor affecting the phenolic content of wines. Different multivariate statistical approaches were used to investigate the relationships between the profile of phenolic compounds and grape variety and also the impact of malolactic fermentation (MLF).
Methods and Results: A reversed phase liquid chromatography/diode array detection method was used for the analysis of major non-flavonoid phenolic compounds in wines from Trincadeira, Aragonez, Cabernet Sauvignon, Alfrocheiro, Casteão and Touriga Nacional varieties before and after MLF. The impact of MLF and grape variety on phenolic profile was evaluated by principal component analysis (PCA), variation partitioning analysis (VPA) and artificial neural network (ANN). PCA explained 86.5% of the total variance among samples. ANN showed a significant clustering of samples according to grape variety and confirmed that MLF has a minor effect on wine phenolic profile. VPA enabled more information to be extracted from the data by identifying explanatory variables responsible for variability among samples.
Conclusions: Compared with PCA and ANN, VPA provides more information concerning the variability on the sample system. Also, grape varieties have a more effective impact on wine low molecular weight phenolic compounds than MLF.
Significance of the Study: Each one of the three multivariate statistical approaches showed ways of analysing large chemistry experimental datasets. VPA is a step forward in data analysis, providing more solid and complete assessment of sample system variability, not possible by PCA and ANN.