Mutation effects predicted from sequence co-variation.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• High-throughput experimental technologies enable assessment of numerous mutations on phenotypes.
• Designing functional assays for these methods is challenging, necessitating robust prediction tools.
• Existing prediction methods often overlook interdependencies between residues or bases.

Purpose of the Study:

• To develop a novel unsupervised statistical method for predicting mutation effects.
• To explicitly capture residue dependencies between positions for improved accuracy.
• To provide a tool for assessing quantitative mutation effects across organisms.

Main Methods:

• Developed EVmutation, an unsupervised statistical method.
• Incorporated explicit modeling of residue dependencies (epistasis).
• Validated predictions against high-throughput mutagenesis experiments and human disease mutation data.

Main Results:

• EVmutation outperforms methods that do not account for epistasis.
• Demonstrated accuracy in predicting mutation effects from experimental and disease data.
• Successfully applied to predict effects for ~7,000 human proteins.

Conclusions:

• EVmutation offers a robust approach for predicting mutation effects by considering residue interdependencies.
• The method enhances the analysis of genetic variation and its phenotypic consequences.
• Pre-computed predictions are available, facilitating broader research applications.