UPicker: a semi-supervised particle picking transformer method for cryo-EM micrographs.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Automatic particle picking is crucial for cryo-electron microscopy (cryo-EM) structure reconstruction.
• Current deep learning methods require extensive manual labeling, leading to biases and suboptimal performance on noisy data.

Purpose of the Study:

• To develop UPicker, a semi-supervised transformer-based method for accurate and robust particle picking in cryo-EM.
• To reduce the need for labor-intensive manual data labeling.

Main Methods:

• UPicker utilizes a two-stage training: unsupervised pretraining with an Adaptive Laplacian of Gaussian (LoG) region proposal generator for pseudo-labeling, followed by supervised fine-tuning.
• Employs contrastive denoising and hybrid data augmentation to improve performance and handle limited labeled data.

Main Results:

• UPicker achieves state-of-the-art accuracy and robustness on simulated and experimental cryo-EM datasets.
• Demonstrates superior performance compared to existing methods, especially with limited labeled data.

Conclusions:

• UPicker offers an effective semi-supervised approach for single particle picking in cryo-EM.
• The method significantly reduces the burden of manual labeling while improving reconstruction quality.