In the long history of Earth, the most serious extinction event occurred at the turn of the Permian Triassic period, approximately 252 million years ago. Professor Song Haijun's team from China University of Geosciences (Wuhan) utilized deep learning technology to reveal how this disaster shaped the evolution of marine life forms. The related results were published on June 11th in Nature Ecology and Evolution. Song Haijun told China Science Daily that morphological diversity, which refers to the differences in morphology among different species, is one of the important indicators for measuring biodiversity. During background extinctions without mass extinction events, species diversity and morphological diversity typically evolve synchronously. Mass extinction events may disrupt this synchronicity, leading to decoupling between the two. The team has developed an automated method called DeepMorph, which combines deep learning models with geometric morphology measurements to extract morphological features from fossil images. Using this method, researchers analyzed fossil images of 599 genera and 6 phyla of marine organisms, revealing the dynamic changes in morphological diversity during the mass extinction period. Research has found that during the Permian Triassic extinction period, most phyla, including ammonites, brachiopods, ostracods, bivalves, and gastropods, experienced selective extinction of complex and decorative forms, with larger, more complex, and strongly ornate forms experiencing more severe extinction, while conodonts showed no signs of selective extinction. Six marine phyla exhibit four different morphological evolution patterns during extinction events. The selectivity and intensity of this extinction may be influenced by complex multiple factors, reflecting differences in environmental tolerance thresholds among different evolutionary branches and morphological types. The research findings provide new insights into how the Great Extinction reshapes biodiversity and ecosystem structure. This study not only deepens people's understanding of the morphological evolution of ancient organisms, but also provides scientific basis for scientists to evaluate the extinction risk faced by modern biodiversity. In addition, DeepMorph and other deep learning methods have demonstrated strong potential in automated and efficient analysis of large-scale data, providing the possibility for further interdisciplinary research between deep learning and geobiology in the future. (Lai Xin She)