A molecular prior distribution for Bayesian inference based on Wilson statistics.

Area of Science:

• Structural biology
• Biophysics
• Computational biology

Background:

• Wilson statistics accurately model protein power spectra at high frequencies, crucial for techniques like cryo-electron microscopy (cryo-EM).
• Recent work provided a rigorous proof of Wilson statistics, revealing correlations in protein scattering potentials.

Purpose of the Study:

• To develop a novel Bayesian prior for molecular structure inference by exploiting statistical estimates of protein scattering potentials.
• To enhance Bayesian inference methods in structural biology, particularly for cryo-EM.

Main Methods:

• Describing the properties and hyperparameter computation of the novel prior.
• Evaluating the prior on synthetic linear inverse problems.
• Comparing the new prior against existing methods in cryo-EM reconstruction across various signal-to-noise ratios (SNRs).

Main Results:

• The novel prior effectively suppresses noise and reconstructs low SNR spectral regions.
• Improved resolution estimates were achieved for tested problems across a wide SNR range.
• Generated Fourier Shell Correlation curves demonstrated robustness against masking effects.

Conclusions:

• The developed prior offers a promising regularization strategy for cryo-EM.
• Potential implications for advancing molecular structure determination using cryo-EM techniques were discussed.