Disentangling the Contribution of Each Descriptive Characteristic of Every Single Mutation to Its Functional Effects

Area of Science:

• Computational Biology
• Bioinformatics
• Protein Engineering

Background:

• Predicting mutational effects is crucial for understanding protein function and engineering new proteins.
• Advanced artificial intelligence (AI) algorithms improve prediction accuracy but lack interpretability.
• Understanding the contribution of specific mutation attributes is the next frontier in this field.

Purpose of the Study:

• To develop an interpretable AI framework to identify contributions of mutation attributes to mutational effects.
• To extract and quantify relationships between physicochemical descriptors and their impact on protein variants.
• To investigate the potential for universal rules in mapping mutation descriptors to effects.

Main Methods:

• Analysis of 29,832 variants from eight deep mutational scan studies.
• Application of an interpretable AI framework to analyze data.
• Extraction of relations between physicochemical descriptors and mutational effect contributions.

Main Results:

• Identified an inverse relationship between fitness/solubility and amino acid distance from catalytic sites.
• Demonstrated universal trends in mutational effect contributions based on Position-Specific Scoring Matrix (PSSM) scores and BLOSUM scores.
• Unsuccessful in explaining quantitative variations in conservation and Solvent Accessible Surface Area (SASA) across different proteins.

Conclusions:

• The interpretable AI framework provides transparency in AI-driven mutational effect predictions.
• Quantified relationships and identified universal trends for specific mutation descriptors.
• Highlights the need for further research to uncover universal rules across diverse proteins.