A triangulation and neural network approach reveals the microbe-metabolite-sensory associations of fermented milk

Understanding the causal associations among microorganisms, flavor metabolites, and sensory attributes is essential for the precise regulation of the flavor of fermented milk. In this study, 22 commercial fermented milks were characterized using sensory evaluation, microbial analysis, and volatile metabolite profiling. Five sensory phenotypes were developed, each exhibiting unique flavor profiles. However, microbial community structures differed only slightly among phenotypes, suggesting that the community structure itself could not fully explain sensory diversification. Therefore, generalized microbe-phenotype triangulation analysis was performed and identified multiple causal taxa associated with different sensory phenotypes, while the mmvec model further revealed strong associations between certain microorganisms and characteristic aroma compounds. Overall, this study established a novel data-driven framework for exploring potential causal associations among microorganisms, flavor metabolites, and sensory phenotypes in fermented milks, representing a preliminary effort to move beyond conventional correlation-based analyses.