Global Analysis of Aggregation Determinants in Small Protein Domains

Area of Science:

• Biochemistry
• Protein Engineering
• Computational Biology

Background:

• Protein aggregation hinders the development of recombinant proteins for biotechnology and therapeutics.
• Predicting protein aggregation is complex, influenced by sequence, structure, environment, and stress.

Purpose of the Study:

• To understand protein aggregation determinants and enhance prediction accuracy.
• To quantify aggregation propensity in small protein domains under stress conditions.

Main Methods:

• Utilized a high-throughput, mass spectrometry-based method to quantify insoluble aggregation in 18,987 small protein domains (40-72 amino acids).
• Applied machine learning to identify sequence and structural features correlated with aggregation.
• Fine-tuned the SaProt protein language model.

Main Results:

• Observed diverse, stress-dependent aggregation phenotypes across protein domains.
• Validated high-throughput mixed-pool aggregation data against individual protein measurements (Pearson's r = 0.65-0.79).
• The fine-tuned SaProt model explained 43-55% of aggregation variation in an independent test set.

Conclusions:

• Identified sequence and structural features influencing protein aggregation.
• The developed model demonstrates potential for engineering aggregation-resistant proteins.
• The generated dataset is a valuable resource for future protein aggregation research.