Discriminatory ability of WL Nutrient Agar for yeasts associated with olive environments: a binary scoring and machine learning approach

Wallerstein Laboratory Nutrient Agar (WLNA) is widely used to differentiate yeast colonies in wine and brewing microbiology, but its discriminatory ability for species associated with table olives and olive oil production environments has never been formally evaluated. We assessed colony features of 227 ascomycetous yeast isolates from these matrices — representing twelve species, predominantly Pichia membranifaciens, P. manshurica, P. kudriavzevii, Wickerhamomyces anomalus, Saccharomyces cerevisiae, Kluyveromyces marxianus, and Candida boidinii — using a binary scoring scheme applied to thirteen characters related to colony colour, shape, appearance and odour and four related to cell shape. Species identities were confirmed by MALDI-TOF mass spectrometry and ITS sequencing (overall agreement: accuracy = 0.90). Initial machine learning classification (CART, Random Forest, SVM-linear, SVM-RBF) on colony features yielded modest performance (accuracy 0.51–0.65), with reliable discrimination limited to S. cerevisiae and Nakazawaea molendinolei. A reproducibility study involving ten operators of varying skill levels identified the root causes: colour and elevation characters contributed disproportionately to inter-operator disagreement while Pichia spp. showed inherently high intraspecific variability. Revised models excluding characters contributing most to disagreement and pooling Pichia species at genus level achieved improved performance (Random Forest: accuracy = 0.72), with acceptable classification of S. cerevisiae, N. molendiolei, and Pichia spp. These results demonstrate that WLNA provides reasonable accuracy in the discrimination of some key species when using robust morphological characters, supporting its use as a low-cost screening tool prior to molecular identification, and establishes a methodological framework that is directly transferable to the evaluation of other differential media.