Bayesian estimation of gene constraint from an evolutionary model with gene features.

Area of Science:

• Genomics
• Population Genetics
• Machine Learning

Background:

• Existing measures of selective constraint on genes are underpowered for short genes, leading to missed pathogenic mutations.
• Accurate constraint metrics are vital for clinical variant interpretation, disease gene discovery, and evolutionary studies.

Approach:

• Developed a novel framework, GeneBayes, integrating a population genetics model with machine learning on gene features.
• Inferred an interpretable constraint metric, shet, designed for improved accuracy across all gene lengths.

Key Points:

• The new shet metric significantly outperforms existing methods in identifying genes crucial for cell essentiality and human disease.
• The framework demonstrates particular strength in detecting constraint in shorter genes, a known limitation of previous approaches.
• GeneBayes offers a flexible platform for estimating various gene-level properties beyond selective constraint.

Conclusions:

• The GeneBayes framework and shet metric provide a powerful tool for characterizing genes relevant to human health and disease.
• Improved constraint estimation will enhance the clinical interpretation of rare coding variants and accelerate disease gene discovery.