Genetic association studies using disease liabilities from deep neural networks

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Traditional case-control studies face limitations with longitudinal health data.
• Prospective cohort studies often struggle with incomplete or changing health outcomes.
• Genome-wide association studies (GWASs) are crucial for understanding genetic traits.

Purpose of the Study:

• To introduce novel methods (liability and meta) for GWASs using deep patient phenotyping.
• To improve the identification of genetic loci associated with binary traits.
• To enhance the predictive power of polygenic risk scores.

Main Methods:

• Developed two new methods, 'liability' and 'meta', for GWASs.
• Utilized deep patient phenotyping to calculate disease liabilities.
• Analyzed 38 common traits in approximately 300,000 UK Biobank participants.

Main Results:

• Identified a greater number of genetic loci compared to conventional case-control GWASs.
• Achieved high replication rates in external GWAS datasets.
• Demonstrated superior prediction of new disease cases using liability-based polygenic risk scores.

Conclusions:

• Integrating high-dimensional phenotypic data with deep neural networks enhances GWASs.
• The proposed methods effectively capture disease-relevant genetic architecture.
• These findings offer a more robust approach to genetic association studies.