Conservation biology requires accurate data on how human-induced threats affect wildlife fitness and survival. Gut microbiota play a critical role in health by influencing physiology, nutrition, immunology, and behaviour. Advances in non-invasive sampling, particularly scat microbiome analysis, offer scalable conservation solutions. This study establishes a benchmark using basic machine learning algorithms (SVM, Ranger, glmnet, and xgboost) to predict health outcomes in koalas from non-invasive scat microbiome data. Scat samples from 125 koalas were analysed using 16S PacBio HiFi sequencing. By incorporating a phylogenetic approach and integrating additional metrics such as sex, age, and stress metabolites, which can potentially be acquired non-invasively, we achieved high accuracy in predicting key health outcomes, including body condition score (BCS), disease status, survival outcome, and weight. The algorithms achieved a minimum accuracy of 68% and a maximum accuracy of 84%. By establishing this benchmark, we set the stage for future research to utilize wildlife hospital infrastructure for larger sample collection and advanced machine learning, with the ultimate goal of developing a predictive health diagnostics tool for wildlife.

Perched high up in a Eucalyptus tree, swaying from side to side, lies a sleepy koala unaware of the means spent each year trying to obtain accurate baseline information about its presence. We have thrown all we could at it, from wildlife surveys to night spotting, bioacoustics, detection dogs, and drones equipped with thermal cameras. Yet, whilst critical to its conservation and management efforts, finding a koala remains an ambitious, time-consuming, and costly endeavour often producing insufficient results. However, little did we know that traces of koalas’ presence and that of its predators along with other native, domesticated, and invasive species, float in the air and can be detected using metagenomics. This study and despite high levels of co-sampled non-target DNA (e.g. humans and domesticated animals) confirms that koalas, species belonging to the wallaby and possum family and threats such as domestic dogs (a major predator contributing to koala population declines) can successfully be detected by sampling airborne particles. Together, it demonstrates the potential of airborne eDNA for the detection of terrestrial wildlife under natural conditions and presents achievable optimisation steps to increase its field applicability and validity.