ZooMR: Automated Taxonomic Identification from Collagen Using a Hierarchical Bayesian Model

Abstract

Zooarchaeology by Mass Spectrometry (ZooMS) is an increasingly important tool for taxonomic identification in archaeological contexts, offering a cost-effective alternative to DNA sequencing. However, current identification methods rely heavily on manual interpretation or heuristic scoring systems, limiting scalability and failing to quantify uncertainty. In this thesis, I develop a hierarchical Bayesian framework for probabilistic taxonomic classification using collagen peptide mass fingerprints. Model parameters are estimated using a combination of empirical data and simple statistical techniques, including binomial likelihoods, logistic regression, and beta-smoothed estimators. Results demonstrate that the framework provides interpretable probability distributions over taxa while addressing key limitations of existing methods. Overall, this work contributes a flexible and extensible statistical approach to ZooMS analysis, with the goal of improving robustness, scalability, and reproducibility in paleoproteomic research.