VMPicker: A novel cryo-EM particle picker leveraging vision mamba and the segment anything model

Cryo-electron microscopy (cryo-EM) has become an essential technique for resolving the three-dimensional structures of biological macromolecules, where accurate particle picking is a critical prerequisite for high-resolution reconstruction. Despite recent advances with convolutional neural networks (CNNs) and Transformer-based models, existing automated methods often encounter difficulties under low signal-to-noise ratio (SNR) conditions and heterogeneous backgrounds. CNNs are constrained in capturing long-range dependencies, whereas Transformers, although effective in modeling global context, tend to underperform in preserving fine structural details and require substantial computational resources. Here, we present VMPicker, a framework integrating Vision Mamba and the Segment Anything Model (SAM). At its core is CryoVMUNet, which extends the standard U-Net by incorporating high-order visual state space (H-VSS) modules alongside convolutional operations and attention mechanisms. This design enables the simultaneous extraction of global dependencies and fine-grained local features while reducing redundant information. By combining CryoVMUNet's high-fidelity predictions with SAM-based automated segmentation, VMPicker achieves precise and robust identification of protein particles in complex cryo-EM micrographs. Benchmarking on ten publicly available cryo-EM datasets shows that VMPicker consistently outperforms state-of-the-art approaches, achieving average improvements of 8.1 % in Precision, 11.2 % in F1 score, and 12.3 % in Dice coefficient, while maintaining competitive Recall. It also produces the highest and most consistent reconstructed density map resolutions across multiple CryoPPP datasets, demonstrating its ability to generate high-quality 3D structures. Furthermore, VMPicker delivers higher computational efficiency with a more compact model design. To our knowledge, this study represents the first application of Vision Mamba to cryo-EM image analysis, underscoring its potential to advance reliable particle picking across diverse experimental conditions.