Genetic background shapes AI-predicted variant effects

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Predicting the functional impact of genetic variants is crucial in biomedicine.
• Current methods often overlook the influence of genetic diversity on variant effects by using a single reference genome.

Purpose of the Study:

• To introduce the personalized variant effect predictor (pVEP) framework.
• To quantify how diverse human genetic backgrounds influence computational predictions of clinical variant effects.

Main Methods:

• Developed the pVEP framework to analyze variant effects across thousands of globally diverse human genomes.
• Applied deep learning models to assess impacts on protein structure, splicing, and noncoding regulation.

Main Results:

• Identified that many clinical variants have heterogeneous predicted effects across different genetic backgrounds (haplotypes).
• Observed that the same variant can be predicted as pathogenic in one genetic context and benign in another.
• Found molecular evidence supporting these findings, including altered protein contacts and splice-site recognition.

Conclusions:

• Personalized genomic context is a critical, yet underappreciated, factor in variant annotation and clinical interpretation.
• This approach has significant implications for understanding genetic variation in diverse populations.