Accurate protein stability predictions from homology models

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein engineering

Background:

• Protein stability changes (ΔΔG) are crucial for predicting effects of amino acid substitutions in protein engineering and interpreting genomic variants for disease risk.
• Structure-based ΔΔG calculations offer high accuracy but are typically limited to experimentally resolved protein structures.
• Many proteins, including a significant portion of the human proteome, lack experimentally determined structures, limiting the application of current ΔΔG prediction tools.

Purpose of the Study:

• To assess the accuracy of ΔΔG values predicted using homology models compared to those derived from experimental crystal structures.
• To determine the minimum sequence identity threshold for reliable ΔΔG predictions on homology models.
• To evaluate the robustness of different ΔΔG calculation methods when applied to homology models.

Main Methods:

• Selected four proteins with extensive experimental ΔΔG data and suitable homology modeling templates across varying sequence identities.
• Employed three distinct computational methods for ΔΔG calculations on both homology models and experimental crystal structures.
• Validated findings by assessing the utility of predicted ΔΔGs in categorizing variant abundance (low vs. wild-type-like).

Main Results:

• ΔΔG values predicted from homology models showed comparable accuracy to those from crystal structures when the template sequence identity was 40% or higher.
• The Rosetta cartesian_ddg protocol demonstrated robustness against structural perturbations introduced by homology modeling.
• An independent assessment confirmed that homology model-based ΔΔGs effectively categorize variant abundance, mirroring results from crystal structure-based calculations.

Conclusions:

• Homology models can reliably substitute for crystal structures in protein stability (ΔΔG) calculations, provided the model is based on a template with at least 40% sequence identity.
• This approach significantly broadens the applicability of accurate ΔΔG predictions to a larger number of proteins, including those not yet structurally resolved.
• The findings support the use of homology modeling for computational predictions in protein engineering and disease variant interpretation.