Robust probabilistic superposition and comparison of protein structures.

Area of Science:

• Structural Bioinformatics
• Computational Biology
• Biophysics

Background:

• Protein structure comparison is crucial in structural bioinformatics.
• Root Mean Square Deviation (RMSD) is the standard measure, but struggles with large conformational changes.
• Optimal superposition is debated for structures with divergent regions, common in NMR ensembles and dynamic proteins.

Purpose of the Study:

• To develop a robust probabilistic method for protein structure superposition and comparison.
• To identify the largest structurally invariant core in proteins with large conformational variations.
• To accommodate non-rigid displacements using outlier-tolerant probability distributions.

Main Methods:

• Modeling non-rigid displacements with heavy-tailed probability distributions.
• Utilizing a scale mixture representation for weighted RMSD calculations.
• Employing Expectation Maximization and Gibbs sampling for parameter estimation and optimal superposition.

Main Results:

• The method effectively handles large-scale conformational changes and divergent structural regions.
• It allows for robust superposition and accurate assessment of NMR structure precision.
• Bayes factors enable automatic model selection, making the method parameter-free.

Conclusions:

• Heavy-tailed distributions are suitable for modeling significant conformational differences in proteins.
• Scale mixture representations facilitate outlier-tolerant superposition and fitting.
• The developed probabilistic method offers robust protein structure comparison, particularly for dynamic systems.