A probabilistic graphical model for estimating selection coefficients of nonsynonymous variants from human population...

Area of Science:

• Genomics
• Population Genetics
• Computational Biology

Background:

• Predicting missense variant effects is crucial for identifying disease risk genes and for clinical genetic diagnostics.
• Current computational methods often predict pathogenicity but lack quantitative human fitness impact.
• Understanding variant effects aids in interpreting genomic data and diagnosing genetic disorders.

Purpose of the Study:

• To develop a novel method, MisFit, for estimating the quantitative missense fitness effect.
• To jointly model molecular and population-level effects of missense variants.
• To improve the accuracy of variant effect prediction beyond simple pathogenicity.

Main Methods:

• Developed MisFit, a graphical model approach.
• Jointly modeled molecular-level effects ($$) and population-level selection coefficients ($$).
• Trained the model using allele counts from 236,017 individuals of European ancestry.

Main Results:

• MisFit's predicted fitness effects ($$) correlate with allele frequencies across ancestries.
• $$ is consistent with de novo mutation fractions in strongly selected sites.
• MisFit outperformed existing methods in prioritizing de novo missense variants in neurodevelopmental disorders.

Conclusions:

• MisFit accurately predicts missense variant fitness effects ($$).
• The method provides novel insights from genomic data, enhancing genetic diagnostics.
• MisFit advances the field of variant effect prediction for disease gene discovery.