Bioinformatic approaches for the genetic and phenotypic characterization of a Saccharomyces cerevisiae wine yeast collection

Abstract

The objective of the present study was to compare genetic and phenotypic variation of 103 Saccharomyces cerevisiae strains isolated from winemaking environments. We used bioinformatics approaches to identify genetically similary strains with specific phenotypes and to estimate a strain's biotechnological potential. A S. cerevisiae collection, comprising 440 strains that were obtained from winemaking environments in Portugal has been constituted during the last years. All strains were genetically characterized by a set of eleven highly polymorphic microsatellites and showed unique allelic combinations. Using neural networks, a subset of 103 genetically most diverse strains was chosen for phenotypic analysis, that included growth in synthetic must media at various temperatures, utilization of carbon sources (glucose, ribose, arabinose, xylose, saccharose, galactose, rafinose, maltose, glycerol, potassium acetate and pyruvic acid), growth in ethanol containing media, evaluation of osmotic and oxidative stress resistance, H2S production and utilization of different nitrogen sources. Using supervised data mining approaches we have found that genotype represented with presence/absence of eleven microsatellites relates well with geographical location (performance evaluation using leave-out-out technique resulted in high performance scores; e.g., area under ROC curve was above 0.8 for a number of standard machine learning approaches tested). To find relations between phenotypes and genotypes, we used a two-step approach which first hierarchically clusters the strains according to their phenotype, and then tests if the resulting sub-clusters are identifiable using strain’s genetic data. Several groups of strains with similar phenotype profiles and common features in genotype were identified this way, and they are subject to further investigations.