CryoEM is a new artificial intelligence method for image analysis: Atelier

To analyze molecular structures using cryo-electron microscope (CryoEM) is a critical part of modern biology research. But interpretation these complex 3D maps faces with limitations of traditional deep learning methods. A new research offers a revolutionary approach to solve these challenges.

Atelier, published in arXiv, the system stands out with the following features:

• The recurrence nerve representations (INR) are using: Instead of traditional fixed votropic gratings, it provides scale independent analysis by modeling volumes with coordinate-based functions
• Hyper networks with efficiency increase: Instead of train a separate INR for each map, it works with multiple maps common learning mechanism using hyper networks
• Providing the spatial localization: Appropriate features depend on certain locations in CryoEM maps, show consistency among examples
• Multi-scale information processing: The system offers more comprehensive analysis by knowledge in different spatial scales
• Payment-based learning: By learning from unlabeled data, it increases data usage efficiency

This study significantly extends the feasibility of deep learning in biological research. The complex nature of cryoEM data was not good fit with the solid structure of traditional nerve networks. Atelier’s coordinate-coached functions approach eliminates this incompatibleity and offers a more flexible, adaptable solution.

The results of the research can start a new era of automatic map attachments in structure biology and related disciplines.