Protein model discrimination using mutational sensitivity derived from deep sequencing.

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Protein structure prediction is crucial for understanding protein function.
• Distinguishing correct protein models from decoys remains a significant challenge in the field.
• Accurate structure-function relationships are vital for drug discovery and protein engineering.

Purpose of the Study:

• To develop a novel computational methodology for accurate protein model selection.
• To establish a rapid and efficient method for assessing mutant protein activities.
• To derive sequence-structure-function relationships from experimental data.

Main Methods:

• Deep sequencing was employed to estimate relative activities of single-site mutants in a bacterial toxin CcdB saturation library.
• A residue-specific empirical score (RankScore) was defined based on phenotypic information.
• The RankScore was correlated with residue depth to identify active-site residues and validate protein models.

Main Results:

• The RankScore methodology successfully identified approximately 98% of correct CcdB models within a 4Å RMSD threshold from a large decoy set.
• The method demonstrated high accuracy in discriminating correct protein structures.
• Validation on eleven additional monomeric proteins confirmed the robustness of the model-discrimination approach.

Conclusions:

• The developed RankScore methodology significantly enhances the accuracy of protein structure prediction by effectively selecting correct models.
• This approach offers a rapid and accurate means to determine relative mutant activities and elucidate sequence-structure-function relationships.
• The methodology is broadly applicable to various biological systems with measurable phenotypic readouts for mutational effects.