Unraveling hidden species diversity of talpid moles using phylogenomics and skull-based deep learning

The sky islands of Southwest China are biodiversity hotspots, where geographic isolation has led to allopatric diversification and cryptic speciation. Here, we integrated phylogenomics, molecular species delimitation, and morphometric analyses to assess the phylogenetic and morphological diversity of the small mammal family of talpid moles. Our findings strongly support recognizing geographically isolated populations as distinct species, highlighting that species diversity in Southwest China’s sky islands is considerably underestimated. As traditional morphology-based methods struggled to detect these cryptic species due to morphological conservatism, we developed a deep learning model that analyzes cranial and mandible images using a hierarchical classification approach, to first differentiate genera and then species. This deep learning model achieved high accuracy (95% genus-level, 90% species-level) with identifying both known and cryptic species. Importantly, it uncovered previously overlooked diagnostic morphological characters thereby demonstrating the potential of deep learning methods to reveal hidden biodiversity within morphologically conserved species complexes, applicable broadly across diverse taxa. Phylogenomics and skull-based deep learning uncover hidden species diversity of talpid moles in the sky islands of southwestern China and show that a hierarchical CNN can identify cryptic species with high accuracy.