Persistent sheaf Laplacian analysis of protein stability and solubility changes upon mutation

Area of Science:

• Computational biology
• Structural bioinformatics
• Topological data analysis

Background:

• Genetic mutations alter protein structure, stability, and solubility, driving diseases.
• Existing computational models lack interpretability and fail to integrate key physicochemical interactions.

Purpose of the Study:

• Introduce SheafLapNet, a novel predictive framework using Topological Deep Learning (TDL) and Persistent Sheaf Laplacian (PSL).
• Enhance the prediction of mutation-induced protein changes by integrating physical and chemical data.

Main Methods:

• Developed SheafLapNet, combining PSL for explicit physicochemical information encoding with protein transformer features.
• Utilized diverse datasets (S2648, S350, S669 for stability; PON-Sol2 for solubility) for rigorous validation.
• Employed multiscale and mechanistic approaches to capture intrinsic molecular interactions.

Main Results:

• SheafLapNet achieved state-of-the-art performance on stability and solubility prediction benchmarks.
• Demonstrated superior accuracy and thermodynamic consistency in predicting mutation effects.
• Showcased enhanced interpretability and generalizability compared to existing methods.

Conclusions:

• Sheaf-theoretic modeling significantly improves the prediction of mutation-induced protein structural and functional changes.
• SheafLapNet offers a more interpretable and generalizable framework for analyzing disease-related molecular alterations.
• The approach provides a powerful tool for understanding the impact of genetic variations on protein properties.