Toward Developing Intuitive Rules for Protein Variant Effect Prediction Using Deep Mutational Scanning Data.

Area of Science:

• Biochemistry
• Computational Biology
• Genetics

Background:

• Single amino acid mutations can significantly alter protein structure and function.
• Artificial intelligence (AI) models predict mutational effects but lack intuitive explanations.
• Existing methods struggle to provide clear insights into the factors driving mutational impact.

Purpose of the Study:

• To develop simple, interpretable thresholding criteria for predicting mutation effects.
• To complement complex AI models with a more intuitive approach to understanding mutational impacts.
• To codify knowledge regarding the consequences of single amino acid substitutions in proteins.

Main Methods:

• Analyzed deep mutational scanning data from all single amino acid substitutions across seven proteins (25,153 mutations).
• Devised thresholding criteria based on five protein descriptors, inspired by Lipinski's rules.
• Defined thresholds using empirical data and evaluated the predictive scope and limitations.

Main Results:

• Established quantitative thresholding rules for predicting mutation effects based on protein descriptors.
• Demonstrated the ability to classify a subset of mutations (neutral or deleterious) with a low error rate.
• Identified the scope and limitations of the developed thresholding criteria.

Conclusions:

• The developed thresholding rules offer a complementary and intuitive approach to AI-based mutation effect predictions.
• These rules facilitate easier interpretation of mutation impacts, particularly for neutral or deleterious classifications.
• The approach aims to enhance the codification of knowledge regarding single amino acid substitution effects on proteins.