Uncovering patterns in skin and gut microbiota of rainbow trout (Oncorhynchus mykiss): insights from machine learning and feeding trials with sustainable aquafeeds based on yellow mealworm (Tenebrio molitor)

Abstract The aquaculture sector has been progressively transitioning toward environmentally sustainable feed production, with insect meals emerging as viable alternatives to fish meal (FM); however, their effects on the microbiota of fish still remain insufficiently characterized. This study examined gut and skin microbiota in rainbow trout ( Oncorhynchus mykiss ) following complete FM substitution with yellow mealworm ( Tenebrio molitor , TM), utilizing machine learning (ML) to investigate diet-microbiota relationships. To this end, microbial abundance data from a prior in vivo trial were analyzed by means of a structured ML pipeline. On the one hand, classification models assessed the association between microbial profiles and dietary regimens, while, on the other hand, regression models evaluated the predictive capacity of feed ingredient variations on microbial abundance shifts. Within this processing framework, feature selection identified informative taxa across taxonomic levels, enhancing model generalizability and reducing overfitting. Several classification algorithms attained optimal accuracy, whereas regression models showed moderate performance, with error values decreasing from higher to lower taxonomic ranks. In particular, feature selection and explainability analyses identified both diet- and tissue-associated indicators: Cutibacterium , Enhydrobacter , and Lactobacillus in the gut; Chryseobacterium , Flectobacillus , and Sphingopyxis in the skin. The occurrence of Deefgea in both tissue types suggested potential water-fish microbial exchange. In conclusion, despite conventional analyses showing only limited dietary modulation, ML models effectively detected diet- and tissue-specific indicators in rainbow trout following FM substitution with TM, ultimately underscoring the potential of integrating AI-driven techniques with next-generation sequencing to uncover ecological patterns across fish tissue types and taxonomic levels.