Expanding the utility of variant effect predictions with phenotype-specific models

Area of Science:

• Genomics
• Computational Biology
• Medical Genetics

Background:

• Current variant effect prediction methods lack specificity for diverse disease outcomes.
• Existing tools often focus on single molecular consequences, limiting clinical utility.
• Differentiating pathogenic variants with varied phenotypic effects is a significant challenge in genomics.

Purpose of the Study:

• To develop a novel machine learning model, Variant-to-Phenotype (V2P), for predicting variant pathogenicity.
• To condition predictions on Human Phenotype Ontology (HPO) disease phenotypes for enhanced accuracy.
• To simultaneously improve variant disease phenotype and effect predictions.

Main Methods:

• Developed V2P, a multi-task, multi-output machine learning model.
• Incorporated disease phenotypes as outputs and during the training process.
• Evaluated V2P against existing variant effect predictors using curated databases and functional assays.

Main Results:

• V2P demonstrates improved prediction of variant pathogenicity and associated disease phenotypes.
• The model successfully identifies pathogenic variants in patient sequencing data.
• V2P outperforms other methods in initial comparisons for variant effect characterization.

Conclusions:

• V2P provides a comprehensive mapping of human genetic variants to disease phenotypes.
• The model's approach enhances the understanding of genotype-phenotype relationships.
• V2P offers a conditioned set of variant effect characterizations for improved genetic diagnostics.