Respiratory microbiota maturation enables machine learning based age prediction in chickens

Abstract Background Understanding how the respiratory microbiota matures with age is key to improving poultry health and pathogen surveillance, yet the ecological processes shaping this transition remain elusive. We aimed to develop an interpretable machine-learning framework capable of identifying age-associated microbial signatures within the chicken nasal microbiota across heterogeneous datasets. Results We compiled data from five independent chicken studies and normalized microbial abundances using Counts Per Million (CPM). To address dataset imbalance and ensure cross-study generalizability, we implemented SMOTE over-sampling and a Leave-One-Study-Out (LOSO) cross-validation framework. Within this architecture, we utilized Recursive Feature Elimination (RFE) to identify a stable consensus signature composed of taxa persisting in at least 70% of the iterations. We benchmarked five algorithms: Classification and Regression Trees (CART), k-nearest neighbors (kNN), Support Vector Machines (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). RF emerged as the best model, achieving a balanced accuracy of 0.965 and a Kappa of 0.920. Consequently, the contribution of each feature was quantified through SHapley Additive exPlanations (SHAP) values on the selected RF model, enabling transparent interpretation of age-dependent microbial patterns. This approach distilled a compact set of predictive taxa, including Corynebacterium, Kocuria, and members of the Micrococcaceae. External validation with longitudinal samples from a Highly Pathogenic Avian Influenza Virus (HPAIV) infection confirmed full generalization, with all 57 samples from 22 chickens correctly classified even under viral-induced conditions. Conclusions The proposed workflow combining LOSO-based feature selection, class-balancing, and interpretable machine learning provides a transferable framework for microbiota-based age inference. Such approaches may inform health monitoring, management of poultry production systems, and wildlife surveillance, illustrating the power of interpretable artificial intelligence to reveal conserved host-microbe dynamics across avian systems. Graphical Abstract