Proteome-wide model for human disease genetics.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Rare Disease Genetics

Background:

• Interpreting missense variants is challenging due to subtle, context-dependent effects.
• Existing prediction models lack proteome-wide calibration, limiting their generalizability.
• Accurate variant deleteriousness prediction is crucial for diagnosing rare genetic disorders.

Purpose of the Study:

• To develop a novel deep generative model, popEVE, for proteome-wide variant deleteriousness estimation.
• To improve the generalizability and calibration of variant effect prediction models.
• To enhance rare disease diagnosis, particularly in singleton cases without parental data.

Main Methods:

• Developed popEVE, a deep generative model integrating evolutionary and human population data.
• Applied popEVE to estimate variant deleteriousness across the human proteome.
• Validated popEVE's performance on a severe developmental disorder cohort.

Main Results:

• popEVE achieves state-of-the-art performance in variant deleteriousness prediction.
• Identified 442 genes with potentially deleterious variants in a developmental disorder cohort, including 123 novel candidates.
• popEVE successfully prioritized causal variants using only child exomes, enabling diagnosis without parental sequencing.

Conclusions:

• popEVE provides a generalizable framework for rare disease variant interpretation.
• Calibrated, evolution-informed scoring models are valuable tools for clinical genomics.
• The model facilitates diagnosis in singleton cases, advancing rare disease genetic interpretation.