Family-specific analysis of variant pathogenicity prediction tools

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Predicting the pathogenicity of missense variants is crucial for understanding genetic diseases.
• Existing prediction tools vary in accuracy and may fail for specific protein families.

Purpose of the Study:

• To benchmark the performance of pathogenicity prediction tools across different protein families.
• To demonstrate a strategy for improving prediction accuracy by tailoring tool selection and thresholds to individual protein families.

Main Methods:

• Performed a protein family-specific benchmarking of multiple missense variant pathogenicity prediction tools.
• Evaluated tool performance using existing annotated missense variant datasets.
• Analyzed functional associations of protein domains with high or low mutation sensitivity.

Main Results:

• All tested tools exhibited high overall accuracy but had specific protein families where predictions were unreliable (accuracy < 51%).
• Selecting optimal tools and thresholds at a protein family level achieved reliable predictions across all Pfam domains (accuracy ≥ 68%).
• Protein domains involved in transcription regulation and DNA binding were highly sensitive to mutations, while immune and stress response domains were less so.

Conclusions:

• Pathogenicity prediction tools require family-specific optimization for reliable variant interpretation.
• Functional annotation of protein domains can enhance the development of future pathogenicity predictors and variant prioritization tools.