Prediction of Protein Aggregation Propensity via Data-Driven Approaches

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Protein aggregation, the clumping of misfolded proteins, is implicated in various amyloid diseases.
• Predicting protein aggregation is crucial for understanding disease mechanisms and developing therapeutics.

Purpose of the Study:

• To develop and compare data-driven surrogate models for predicting protein aggregation.
• To evaluate the performance of feature-based and graph-based models using different databases.

Main Methods:

• Constructed an aggregation propensity score database using Aggrescan3D 2.0 and Protein Data Bank structures.
• Developed feature-based and graph-based models for protein aggregation prediction.
• Built a feature-based model using the Curated Protein Aggregation Database 2.0 for experimental data.

Main Results:

• The graph-based model, requiring protein structures, achieved an R² of 0.95, outperforming the feature-based model.
• A feature-based model effectively predicted aggregated intensity curves from experimental data.

Conclusions:

• The study presents effective data-driven approaches for predicting protein aggregation.
• Model performance is dependent on the chosen descriptors and database type.