Machine Learning Techniques Disclose the Combined Effect of Fermentation Conditions on Yeast Mixed-Culture Dynamics and Wine Quality.

The use of yeast starter cultures consisting of a blend of and non- yeasts has increased in recent years as a mean to address consumers' demands for diversified wines. However, this strategy is currently limited by the lack of a comprehensive knowledge regarding the factors that determine the balance between the yeast-yeast interactions and their responses triggered in complex environments. Our previous studies demonstrated that the strain UTAD222 has potential to be used as an adjunct of in the wine industry due to its positive impact on the fruity and floral character of wines. To rationalize the use of this yeast consortium, this study aims to understand the influence of production factors such as sugar and nitrogen levels, fermentation temperature, and the level of co-inoculation of UTAD222 in shaping fermentation and wine composition. For that purpose, a Central Composite experimental Design was applied to investigate the combined effects of the four factors on fermentation parameters and metabolites produced. The patterns of variation of the response variables were analyzed using machine learning methods, to describe their clustered behavior and model the evolution of each cluster depending on the experimental conditions. The innovative data analysis methodology adopted goes beyond the traditional univariate approach, being able to incorporate the modularity, heterogeneity, and hierarchy inherent to metabolic systems. In this line, this study provides preliminary data and insights, enabling the development of innovative strategies to increase the aromatic and fermentative potential of UTAD222 by modulating temperature and the availability of nitrogen and/or sugars in the medium. Furthermore, the strategy followed gathered knowledge to guide the rational development of mixed blends that can be used to obtain a particular wine style, as a function of fermentation conditions.