Discovering functionally important sites in proteins.

Area of Science:

• Biochemistry
• Computational Biology
• Genetics

Background:

• Missense variants are a major cause of genetic diseases.
• Identifying functionally important protein sites is challenging due to limited data and confounding factors like structural stability.
• Existing methods struggle to distinguish functional importance from structural constraints.

Purpose of the Study:

• To develop a machine learning model for predicting functional sites in proteins.
• To integrate sequence conservation with biophysical stability models for accurate prediction.
• To validate the model's utility in discovering active, regulatory, and binding sites and understanding disease mechanisms.

Main Methods:

• Developed a machine learning method combining statistical protein sequence models with biophysical stability models.
• Trained the model using multiplexed experimental data on variant effects.
• Validated the model broadly across various protein functions and in a specific disease context.

Main Results:

• The model successfully predicts functional sites, including active, regulatory, and binding sites.
• Demonstrated the model's ability to pinpoint molecular mechanisms of disease-causing missense variants.
• Prospective prediction and experimental validation confirmed the model's utility for HPRT1 variants linked to Lesch-Nyhan syndrome.

Conclusions:

• The novel machine learning approach effectively identifies functionally important protein sites.
• This method advances the understanding of protein function, variant effects, and disease pathogenesis.
• The tool has significant implications for biotechnology, pharmacology, and genetic disease research.