Detecting continuous structural heterogeneity in single molecule localization microscopy data with a point cloud...

Area of Science:

• Biophysics
• Structural Biology
• Computational Biology

Background:

• Single molecule localization microscopy (SMLM) struggles with low signal-to-noise ratios for macromolecular complexes due to limited labeling and photon counts.
• Particle fusion enhances SMLM image quality but assumes structural homogeneity, which is often not the case.
• Heterogeneity in SMLM data can stem from geometric variations or different conformational states of the structures.

Purpose of the Study:

• To develop a computational method for detecting multiple modes of variation in SMLM datasets.
• To analyze 2D and 3D localization data directly, bypassing the need for pixelated images.
• To quantify nanoscale structural heterogeneity in macromolecular complexes.

Main Methods:

• Introduction of a Point Cloud Variational Auto-Encoder (PC-VAE) designed for 2D and 3D localization data.
• The PC-VAE operates directly on lists of localizations, offering computational efficiency.
• Linear scaling with dataset size and rapid network training (four epochs) enable analysis in minutes.

Main Results:

• The PC-VAE successfully detected nanoscale radius variations in 2D Nuclear Pore Complex data.
• Height variations were identified in 3D DNA origami tetrahedron datasets.
• Both radius and height variations, on the few-nanometer scale, were detected in 3D Nuclear Pore Complex data.

Conclusions:

• The developed PC-VAE method effectively detects subtle, nanoscale structural heterogeneity in macromolecular complexes using SMLM data.
• The computational approach provides a rapid and scalable solution for analyzing complex biological structures.
• This method advances the capability to characterize conformational flexibility and geometric variations in biological assemblies.