Cobb Savannah E, La Darrell, Cooke Siobhán B
Johns Hopkins University School of Medicine, Baltimore, Maryland, USA.
Johns Hopkins University, Baltimore, Maryland, USA.
J Morphol. 2025 Aug;286(8):e70068. doi: 10.1002/jmor.70068.
Reconstructing the diets of extinct taxa is essential for understanding their ecologies and evolutionary histories, yet traditional methods and proxies such as molar morphology have limited resolution. The potential of premolar morphology as a dietary proxy remains underexplored, and advanced computational methods have rarely been applied to improve dietary inference in paleontology. We integrate Random Forest (RF) machine learning and comparative phylogenetic methods to identify and rank dental proxies for diet in a large sample of anthropoid primates. We quantify dietary trends in premolar topography and cusp relief and find that premolar protoconid relief is a strong predictor of dietary category, especially for distinguishing hard-object feeders, which outperformed traditional proxies on molars and incisors. We also identify sexually dimorphic dietary trends in honing premolars. Feature selection improved classification accuracy by 5%-11% compared to unpruned models, with the highest accuracy achieved by a model incorporating premolar, molar, and incisor data. These findings establish robust new dental proxies for dietary inference and demonstrate the potential of machine learning and a multi-tooth approach in ecomorphological research. By expanding the toolkit for reconstructing the diets of extinct primates, we establish a framework that may help clarify the ecological pressures that have shaped the evolution of modern clades including that of the human lineage.
重建已灭绝类群的饮食结构对于理解它们的生态和进化历史至关重要，然而传统方法和指标，如臼齿形态，分辨率有限。前臼齿形态作为饮食指标的潜力仍未得到充分探索，先进的计算方法也很少应用于改进古生物学中的饮食推断。我们整合随机森林（RF）机器学习和比较系统发育方法，以识别和排序大量类人猿灵长类样本中用于饮食推断的牙齿指标。我们量化了前臼齿地形和牙尖起伏的饮食趋势，发现前臼齿原尖起伏是饮食类别的有力预测指标，特别是在区分硬食性动物方面，其表现优于传统的臼齿和门齿指标。我们还识别了磨蚀性前臼齿中的两性异形饮食趋势。与未修剪的模型相比，特征选择将分类准确率提高了5%-11%，包含前臼齿、臼齿和门齿数据的模型准确率最高。这些发现为饮食推断建立了可靠的新牙齿指标，并证明了机器学习和多齿方法在生态形态学研究中的潜力。通过扩展重建已灭绝灵长类饮食结构的工具包，我们建立了一个框架，这可能有助于阐明塑造包括人类谱系在内的现代类群进化的生态压力。
