EPicker is an exemplar-based continual learning approach for knowledge accumulation in cryoEM particle picking.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Deep learning is crucial for automated particle picking in single-particle cryo-electron microscopy (cryo-EM).
• Existing supervised methods struggle with generalization to new datasets and retaining knowledge from older data.
• Catastrophic forgetting limits the continuous improvement of deep learning models in cryo-EM particle picking.

Purpose of the Study:

• To develop a novel deep learning approach for robust and adaptable particle picking in cryo-EM.
• To enable continuous learning and knowledge accumulation without forgetting previously learned information.
• To expand the utility of automated particle picking beyond protein particles to other biological objects.

Main Methods:

• An exemplar-based continual learning strategy was implemented in a program named EPicker.
• The approach trains existing models on new samples, integrating new knowledge while preventing catastrophic forgetting.
• The model was designed to identify not only protein particles but also general biological objects like vesicles and fibers.

Main Results:

• EPicker demonstrates the ability to accumulate knowledge from new datasets without losing previously acquired information.
• The program shows improved generalization performance on unseen datasets with different features.
• EPicker successfully identifies a broader range of biological objects, including vesicles and fibers, in addition to protein particles.

Conclusions:

• Exemplar-based continual learning offers a solution to the limitations of traditional supervised methods in cryo-EM particle picking.
• EPicker provides a continuously adaptable tool for automated particle identification, enhancing the efficiency of cryo-EM data processing.
• The developed method broadens the scope of deep learning applications in structural biology and biophysics by enabling the identification of diverse biological structures.