Distribution free methods to model the content of biogenic amines in Spanish wines

• Non-parametric approach for the interpretation of chemical data when normality fails
• β-content tolerance interval guarantees the proportion of wines that meet a limit.
• A confidence region is obtained with Clayton copula and Gaussian kernel margins.
• 684 samples of Spanish wines are analyzed to determine biogenic amines content.
• High amounts of biogenic amines in wine can lead to health problems

Biogenic amines are formed from precursor amino acids by various microorganisms present in wine, and this may happen at any step of production, ageing or storage. The presence of these compounds is important because high amounts of them can lead to health problems. Also, biogenic amines can be potentially applied as indicators of food spoilage and/or authenticity.Nevertheless, there are complex relationships among biogenic amines in Spanish wines and their distribution is far from a normal distribution. Even though, this structure must be taken into account to provide a frame of reference on the quality of wines and to contribute to the efforts of the entire productive chain to attain consumer safety.In this work, 684 samples of wines from different Spanish regions have been analyzed in order to determine the content of histamine, tyramine, phenylethylamine, cadaverine and putrescine during 2010, 2014 and 2015. The statistical distribution of histamine has been modelled by using the β-content tolerance intervals. Besides, copulas to obtain the joint multivariate confidence region between histamine and tyramine have been built for the first time in the oenological field. The β-content tolerance intervals for histamine content, in the case of the population of red wines in 2010, lead to decide that only 53.9% of the distribution would be below 10 ppm (with a 95% confidence in this affirmation). This percentage rises from 53.9 to 74.6% in the year 2014 and to 90.2% in 2015. Besides, the conjoint distribution of histamine and tyramine is modeled by a Clayton copula with margins estimated by Gaussian kernel, which allows concluding that this distribution is similar for the three years.