DeepSVP: integration of genotype and phenotype for structural variant prioritization using deep learning.

Area of Science:

• Genomics
• Computational Biology
• Medical Genetics

Background:

• Structural genomic variants contribute significantly to human variability and are implicated in various diseases.
• Interpreting the phenotypic impact of structural variants is challenging due to their complexity and potential to affect multiple genes.
• Existing phenotype-based methods struggle with the intricate nature of structural variants compared to single nucleotide variants.

Purpose of the Study:

• To develop a computational method for prioritizing structural variants associated with genetic diseases.
• To integrate diverse biological data, including gene functions, expression, and phenotypic consequences, for variant interpretation.

Main Methods:

• Developed DeepSVP, a computational tool that combines genomic information with gene functions and phenotype data.
• Utilized ontologies and machine learning to systematically link gene product functions and expression patterns to phenotypic outcomes.
• Incorporated gene expression data across cell types and anatomical sites.

Main Results:

• DeepSVP significantly enhances the success rate of identifying causative variants in benchmark datasets.
• The method demonstrates effectiveness in pinpointing novel pathogenic structural variants, particularly in consanguineous families.
• Achieved improved prioritization of structural variants involved in genetic disorders.

Conclusions:

• DeepSVP offers a robust computational approach for prioritizing disease-associated structural variants.
• The integration of multi-modal data, including functional genomics and phenotypic information, is crucial for understanding variant impact.
• This tool advances the identification of genetic variants underlying human diseases.