Revealing Ancient Wheat Phylogenetic Diversity: Machine Learning and Logistic Regression Identify Triticum...

Identifying archaeobotanical wheat remains is central to reconstructing the evolutionary history of cereal crops. Beyond documenting agricultural practices, such analyses provide critical evidence of phylogenetic diversity, lineage persistence, and local extinction events within the genus Triticum L. This study applies advanced computational morphometrics to reveal deep-time changes in wheat species distribution, including the disappearance of taxa now phylogeographically confined to central Asia.

We developed a machine learning framework integrating Random Forest compared with logistic regression to classify morphometric data from 848 dry and 340 experimentally carbonized modern grains representing multiple wheat taxa (genus Triticum), alongside 15 archaeobotanical T. turgidum subsp. parvicoccum and 38 T. aestivum var. antiquorum. This probabilistic classifier was then applied to 2463 archeological wheat grains, including 48 from Punta de los Gavilanes and 517 from Almizaraque (southeastern Spain, 3rd-2nd millennium BC).

The analysis identified Triticum sphaerococcum and other phylogenetically distinct wheat taxa-today restricted to central and south Asia-among western European Bronze Age assemblages. These findings indicate that lineages now regionally extinct once formed part of a broader cultivated gene pool spanning into the western Mediterranean. Morphometric evidence highlights that past wheat diversity encompassed multiple clades and morphotypes absent from modern European germplasm.

Our results demonstrate substantial phylogenetic turnover in wheat over the past 4000 years, marked by regional extirpations and contraction of once-widespread lineages to central Asia. This provides rare archeological evidence for the tempo and mode of cereal phylogeography, illustrating how domesticated lineages underwent extinction and range restriction akin to wild taxa. By integrating computational morphometrics with archaeobotanical evidence, this study establishes a scalable framework for tracing cryptic phylogenetic diversity, refining models of wheat domestication and assessing long-term genetic erosion in cultivated plants.