Combining structural modeling with ensemble machine learning to accurately predict protein fold stability and binding...

Area of Science:

• Computational biology
• Protein engineering
• Genomics

Background:

• Sequencing technologies have rapidly identified numerous mutations, but understanding their functional and disease implications requires biochemical insights.
• Predicting the impact of coding mutations on protein stability and binding affinity is crucial for mechanistic studies.
• Existing methods rely on semi-empirical force fields or machine learning, each with limitations.

Purpose of the Study:

• To develop a highly accurate computational method for predicting the effects of mutations on protein stability and binding affinity.
• To integrate diverse data sources, including structural and sequence features, for improved prediction.
• To enable large-scale, proteome-wide predictions of mutation effects.

Main Methods:

• Introduction of ELASPIC, an ensemble machine learning approach combining semi-empirical energy terms, sequence conservation, and molecular details.
• Utilizing a Stochastic Gradient Boosting of Decision Trees (SGB-DT) algorithm for prediction.
• Integration of homology modeling to facilitate proteome-wide predictions on modeled protein structures.

Main Results:

• ELASPIC significantly outperforms existing methods, achieving correlation coefficients of 0.77 for stability and 0.75 for affinity predictions.
• Accurate predictions are possible even on protein structures generated through homology modeling.
• The method identified distinct patterns for disease-associated mutations compared to neutral mutations.

Conclusions:

• The combined approach in ELASPIC offers a substantial improvement in predicting mutation effects.
• ELASPIC provides molecular-level insights into protein instability caused by mutations, aiding disease mechanism understanding.
• The tool facilitates proteome-wide analysis, distinguishing between disease-related and benign mutations.