### Abstract Recent advancements in artificial intelligence (AI) have significantly empowered the field of in situ structural biology, particularly in the challenging task of particle picking in cryo-electron tomography (Cryo-ET). The core events of this development involve the collaboration between the Multimodal Artificial Intelligence System Laboratory at the Institute of Automation, Chinese Academy of Sciences (CAS), led by Dr. Ge Yang, and the Biological Imaging Center at the Institute of Biophysics, CAS, led by Dr. Fei Sun. Together, they have introduced DeepETPicker, a novel method based on weakly supervised deep learning, which aims to enhance the efficiency and accuracy of particle picking in Cryo-ET. **Key Events:** 1. **Development of DeepETPicker:** The team developed DeepETPicker, a method that requires only a minimal amount of manual annotation to train, thereby significantly reducing the time and effort needed for particle picking. 2. **Technical Innovations:** To address the limitations of existing methods, DeepETPicker employs several innovative techniques: - **Simplified Labels:** The method uses weak labels (TBall-M) instead of ground-truth masks, reducing the burden of manual annotation. - **Efficient Model Architecture:** DeepETPicker incorporates coordinate convolution and image pyramid inputs into a 3D-ResUNet segmentation architecture, enhancing the accuracy of particle localization. - **Data Augmentation and Overlap Partitioning:** Rich data augmentation techniques and an overlap partitioning strategy are used to improve model performance with small training sets. - **GPU-Accelerated Post-Processing:** The method utilizes a GPU-accelerated mean pooling and non-maximum suppression (MP-NMS) operation, which speeds up the particle center localization process by several orders of magnitude compared to clustering-based methods. 3. **User-Friendly Software:** An open-source software with a simple and intuitive user interface has been developed to assist users in image preprocessing, particle annotation, model training, and inference. 4. **Performance Evaluation:** DeepETPicker was benchmarked against the current state-of-the-art particle picking methods on multiple Cryo-ET datasets. The evaluation metrics included precision-recall, F1-score, log-likelihood contribution, maximum value probability, Rosenthal-Henderson resolution, and global resolution. The results demonstrated that DeepETPicker outperforms existing methods, achieving both high speed and accuracy, and producing structural reconstructions of biological macromolecules that match the resolution obtained through expert manual picking. **People:** - **First Authors:** Guole Liu (Assistant Researcher, Institute of Automation, CAS) and Tongxin Niu (Engineer, Institute of Biophysics, CAS). - **Corresponding Authors:** Ge Yang (Researcher, Multimodal Artificial Intelligence System Laboratory, CAS) and Fei Sun (Researcher, State Key Laboratory of Biomacromolecules, Institute of Biophysics, CAS). - **Other Contributors:** Mengxuan Qiu (Ph.D. student, Ge Yang's group) and Yun Zhu (Researcher, Fei Sun's group). **Location:** - **Institutes:** Institute of Automation, CAS, and Institute of Biophysics, CAS, both located in China. **Time:** - **Publication Date:** The research was recently published in *Nature Communications*. **Significance:** The introduction of DeepETPicker marks a significant step forward in in situ structural biology, particularly in the context of Cryo-ET. By reducing the dependency on extensive manual annotations and optimizing computational efficiency, this method promises to streamline the process of particle picking, making high-resolution structural analysis more accessible and feasible for researchers. The open-source software further enhances the practical utility of DeepETPicker, potentially accelerating advancements in the field of structural biology and contributing to a deeper understanding of the dynamic behavior of biological macromolecules in their natural cellular environments. **Support and Funding:** The research was supported by several prestigious grants, including the CAS Strategic Priority Research Program (B class), the National Natural Science Foundation of China, the National Key Research and Development Program, CAS, and the University of Chinese Academy of Sciences. Additionally, the technology has been granted a Chinese patent. **Links:** - **Paper:** [DeepETPicker: Fast and accurate 3D particle picking for cryo-electron tomography using weakly supervised deep learning](https://doi.org/10.1038/s41467-024-46041-0) - **Code:** [DeepETPicker GitHub Repository](https://github.com/cbmi-group/DeepETPicker) This abstract provides a concise overview of the key elements of the news article, focusing on the main events, technical innovations, performance evaluation, and the significance of the research.