Accurate assignment of disease liability to genetic variants using only population data

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Public variant repositories are growing, necessitating accurate pathogenicity prediction methods.
• Existing methods for DNA variant pathogenicity prediction require evaluation.

Purpose of the Study:

• To test the accuracy of DNA variant pathogenicity prediction using only population data.
• To develop and validate a novel Bayesian method for variant pathogenicity assessment.

Main Methods:

• Developed the Bayesian prevalence ratio (BayPR) method based on variant prevalence in healthy versus affected populations.
• Assigned probabilities of pathogenicity using a Bayesian approach assuming two distinct variant distributions (pathogenic and benign).
• Validated BayPR using expertly curated variants in CFTR and genes associated with Mendelian conditions.

Main Results:

• BayPR accurately classified 300 of 313 CFTR variants (95.8%), distinguishing pathogenic and benign variants.
• Achieved an area under the receiver operating characteristic curve of 0.99 for missense CFTR variants, outperforming 10 common algorithms.
• Assigned high disease-causing probabilities (≥80%) to 98.3% of pathogenic variants and low probabilities (≤20%) to 95.7% of benign variants across 8 Mendelian disease genes.

Conclusions:

• The BayPR approach accurately predicts DNA variant pathogenicity using only population prevalence data.
• This method is effective for variants causing Mendelian disorders, irrespective of variant type or functional effect.
• BayPR offers a robust tool for interpreting genetic variants in large-scale population studies.