DeepGenePrior: A deep learning model for prioritizing genes affected by copy number variants.

Area of Science:

• Neurogenetics
• Computational Biology
• Genomics

Background:

• Genetic brain disorders exhibit high heterogeneity, complicating gene discovery.
• Current gene prioritization methods rely on limited evidence and can yield false positives/negatives.
• Identifying causative genes is crucial for understanding and treating central nervous system abnormalities.

Purpose of the Study:

• To introduce DeepGenePrior, a deep neural network model for prioritizing candidate genes in genetic brain disorders.
• To develop a novel scoring system using Variational AutoEncoder (VAE) for gene impact assessment.
• To exclusively utilize copy number variants (CNVs) for gene prioritization, overcoming limitations of existing methods.

Main Methods:

• Developed DeepGenePrior, a deep learning model leveraging Variational AutoEncoder (VAE).
• Analyzed CNV data from 74,811 individuals across autism, schizophrenia, and developmental delay cohorts.
• Prioritized candidate genes based solely on CNV data, without relying on prior associations or auxiliary data.

Main Results:

• Achieved a 12% increase in fold enrichment for brain-expressed genes compared to prior studies.
• Observed a 15% increase in genes linked to mouse nervous system phenotypes.
• Identified shared deletions in ZDHHC8, DGCR5, and CATG00000022283 across all three disorders, suggesting common genetic underpinnings.

Conclusions:

• DeepGenePrior effectively prioritizes candidate genes for brain disorders using CNVs.
• The findings suggest a potential shared genetic etiology for autism, schizophrenia, and developmental delay.
• The DeepGenePrior model is publicly available to advance gene discovery in complex neurological conditions.