Computational assessment of feature combinations for pathogenic variant prediction

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Predicting genetic variant effects and disease association remains challenging in sequencing studies.
• Accurate identification and prioritization of pathogenic variants are crucial for genetic research and clinical applications.

Purpose of the Study:

• To compare various variant and gene-specific features for predicting pathogenic variants.
• To investigate the optimal combination of features and existing methods for improved prediction performance.

Main Methods:

• Comparative analysis of different variant and gene-specific features.
• Evaluation of existing pathogenicity prediction tools (e.g., PON-P2, PROVEAN).
• Investigating the synergistic effects of combining features and methods.

Main Results:

• A combination of Gene Ontology (GO) annotations (biological process) with PON-P2 and PROVEAN significantly enhances pathogenic variant prediction.
• The effectiveness of GO annotations is attributed to the multifunctional nature of disease genes, not variant-specific bias.
• A subset of challenging variants was identified where current prediction methods exhibit limitations.

Conclusions:

• Existing methods for predicting variant pathogenicity can be substantially improved.
• Integrating gene-level functional information with computational prediction tools offers a promising avenue for enhanced variant interpretation.