Machine Learning Species Identification with ZooMS Collagen Fingerprinting
The creation of a robust method of species identification using collagen fingerprinting, also known as ZooMS (Zooarchaeology by Mass Spectrometry) has been useful for objectively defining the composition of the fragmentary component of archaeological assemblages. The method usually works through the measurements of the sizes of collagen peptides following enzymatic digestion, which yield a fingerprint that can be genus or even species-specific. However, even these peptide biomarkers have been determined with a level of subjectivity, largely influenced by those most peaks that are commonly observed. In some taxa, particularly the domesticates, sequence data can be used to confirm this variation but ZooMS is being increasingly applied to a greater range of wild taxa. In the absence of sequence information, peptide biomarkers are proposed through observations from reference material, but often with less confidence. Here we apply machine learning to our assemblage of >10,000 Pin Hole Cave (Creswell Crags, UK) fragmentary Late Pleistocene microfaunal remains, demonstrating that an objective manner of biomarker determination can be achieved which builds upon those already determined manually. Most importantly, machine learning approaches appear able to rapidly identify not only known species but can reveal taxa beyond the expected, widening our knowledge of a faunal assemblage.
Cite this Record
Machine Learning Species Identification with ZooMS Collagen Fingerprinting. Michael Buckley, Muxin Gu. Presented at The 82nd Annual Meeting of the Society for American Archaeology, Washington, DC. 2018 ( tDAR id: 443058)
This Resource is Part of the Following Collections
min long: -11.074; min lat: 37.44 ; max long: 50.098; max lat: 70.845 ;
Abstract Id(s): 21700