DiffFit: Visually-Guided Differentiable Fitting of Molecule Structures to a Cryo-EM Map.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Cryo-electron microscopy (cryo-EM) generates 3D density maps of molecular structures.
• Fitting atomic models into cryo-EM maps is crucial for understanding protein function.
• Current methods often require manual intervention and are time-consuming.

Purpose of the Study:

• To develop an automated and accurate method for fitting protein structures into cryo-EM data.
• To improve the visualization and revision process for structural biologists.
• To enable faster integration of known and predicted protein structures.

Main Methods:

• Introduced DiffFit, a differentiable algorithm for automated structure fitting.
• Utilized differentiable 3D rigid transformations of atomic coordinates.
• Employed a novel loss function based on multi-resolution arrays and negative space exploitation.
• Sampled density values from cryo-EM maps at atomic coordinates.

Main Results:

• DiffFit demonstrated superior placement quality compared to existing methods on large datasets.
• The algorithm enables automatic fitting and interactive revision of results.
• Successfully applied to integrate known structures and fit predicted domains.

Conclusions:

• DiffFit offers an automated, accurate, and efficient solution for fitting protein structures into cryo-EM maps.
• The method enhances the utility of cryo-EM data for structural biology.
• The open-source plugin facilitates broader adoption and research advancement.