Machine-learning of complex evolutionary signals improves classification of SNVs.

Area of Science:

• Genomics
• Bioinformatics
• Evolutionary Biology

Background:

• Conservation of genetic sequences across species is a key indicator of variant pathogenicity.
• Complex conservation patterns in some genomic regions challenge traditional prediction models.
• Understanding species-specific conservation is crucial for accurate variant interpretation.

Purpose of the Study:

• To investigate the association between complex conservation patterns and single-nucleotide variant (SNV) pathogenicity.
• To evaluate the impact of species and gene choice on conservation-based variant prediction accuracy.
• To develop an improved method for predicting SNV pathogenicity using optimized conservation analysis.

Main Methods:

• Analysis of SNV pathogenicity and conservation patterns in 115 disease genes across 99 vertebrate species.
• Development of EvoDiagnostics, a random-forest machine-learning model utilizing species-specific conservation as a feature.
• Comparative performance evaluation against traditional and deep-learning based prediction algorithms.

Main Results:

• Conservation accuracy for variant pathogenicity is highly dependent on the specific set of species and genes analyzed.
• Certain genes exhibit species-specific conservation patterns that optimize prediction accuracy.
• EvoDiagnostics demonstrated superior performance in predicting variant pathogenicity compared to existing methods.

Conclusions:

• Conservation is not a universal predictor; its utility varies significantly by species and gene context.
• Optimizing conservation analysis on a per-species and per-gene basis enhances the prediction of variant pathogenicity.
• EvoDiagnostics offers a more biologically relevant and accurate approach for variant pathogenicity prediction.