Graph identification of proteins in tomograms (GRIP-Tomo) 2.0: Topologically aware classification for proteins

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Cryo-electron tomography (cryo-ET) provides near-native biomolecular structures.
• Current methods for analyzing cryo-ET data are computationally intensive and struggle with small proteins.
• Identifying proteins within noisy tomographic volumes is a significant challenge.

Purpose of the Study:

• To develop a faster, more accurate method for identifying protein structures in cryo-ET data.
• To overcome limitations of existing computational approaches for protein detection.
• To advance automated protein identification in cryo-ET workflows.

Main Methods:

• Utilized a Graph Network and topologically invariant approach for protein identification.
• Developed GRIP-Tomo 2.0, a machine-learning pipeline extracting topological features.
• Incorporated synthetic tomogram generation, graph-based feature extraction, and High Performance Computing acceleration.

Main Results:

• GRIP-Tomo 2.0 achieves over 90% accuracy distinguishing proteins from noise in synthetic datasets.
• Demonstrated over 80% accuracy on real cryo-ET datasets with high resolution (approx. 1 Å/pixel).
• Successfully identifies protein density amidst noisy experimental backgrounds.

Conclusions:

• GRIP-Tomo 2.0 offers a significant improvement for analyzing cryo-ET data.
• The algorithm facilitates the automated detection of both small and large proteins.
• This work is a foundational step towards enhanced visual proteomics and automated cryo-ET analysis.