Rumen Microbiota-Based Machine Learning Approach for Predicting Heat Stress and Identifying Associated Microbes.

Heat stress poses a significant global challenge to sustainable livestock production, leading to detrimental impacts on animal production and welfare. Reduced appetite and increased body temperature further disrupt the gastrointestinal microbial ecosystem of heat-stressed animals, altering nutrient digestion and affecting host production. However, reported heat-stress-associated microbes have varied across studies, partly due to inconsistencies in microbiota analysis pipelines and taxonomic levels reported. In this study, to identify consistent rumen microbial taxa influenced by heat stress and evaluate potential of rumen microbiota in heat stress prediction, we collected publicly available raw 16S rRNA gene amplicon sequencing data of rumen fluid samples from lactating Holstein cattle housed in thermoneutral or heat stress condition from eight studies, analyzed their microbial composition using a consistent bioinformatic pipeline, and built machine learning models with the rumen microbiota profile to predict heat stress. Important rumen microbial taxa were selected using Boruta (a feature selection algorithm to identify important features) as potential biomarkers to predict heat stress, such as lactate-producing bacteria Lactobacillales, fiber-degrading bacteria Ruminococcaceae UCG-001, and methanogenic archaea Methanomicrobium. Additionally, the random forest model using the available animal factors and relative abundance of rumen microbial taxa showed a much higher performance for heat stress prediction, compared to the model without rumen microbiota profile (Area Under the Curve: 0.851 vs. 0.440). This study confirmed a distinct rumen microbiota signature in heat-stressed lactating Holstein cattle and identified specific rumen microbial taxa as potential biomarkers that could be targeted to mitigate heat-stress responses in dairy cows.