Bayesian inference of protein structure from chemical shift data.

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Protein chemical shifts are crucial for refining molecular mechanics force fields in simulations.
• Empirically determined weights for chemical shift restraints can introduce bias and lead to inaccurate protein structures.

Purpose of the Study:

• To develop an inferential structure determination framework that jointly models protein structure and chemical shift prediction errors.
• To compare the accuracy of structures obtained using Gaussian or Cauchy distributions for chemical shift errors against empirically weighted potentials.

Main Methods:

• Formulation of a joint probability distribution for structure and chemical shift prediction errors.
• Markov chain Monte Carlo simulations of three small proteins (ENHD, Protein G, SMN Tudor Domain) using PROFASI and CamShift.
• Utilized clustering criteria and solvent exposure scoring for structure identification.

Main Results:

• Sampling both protein structure and chemical shift prediction uncertainties yielded more accurate structures than conventional empirical weighting.
• The Cauchy distribution, whether using sampled uncertainties or predetermined weights, demonstrated superior convergence to the native protein fold.
• The study highlights the potential of probabilistic modeling of prediction errors in structural biology.

Conclusions:

• The proposed inferential framework improves protein structure prediction accuracy by incorporating chemical shift prediction uncertainties.
• Both Gaussian and Cauchy distributions offer advantages, with Cauchy distributions showing better convergence to native structures.
• This work advances computational approaches for determining protein structures by leveraging full data information content.