Variational autoencoder-based model improves polygenic prediction in blood cell traits.

Area of Science:

• Genomics
• Computational Biology
• Personalized Medicine

Background:

• Large-scale genomic studies enable genetic prediction of complex traits.
• Polygenic risk scores (PRSs) aggregate genomic information for personalized risk prediction.
• Conventional linear PRS models struggle with high-dimensional genomic data and interaction effects.

Purpose of the Study:

• To enhance the predictive power of PRSs using advanced deep learning techniques.
• To develop a novel deep learning-based PRS construction method.
• To improve the accuracy of genetic predisposition assessment for complex traits.

Main Methods:

• Application of a variational autoencoder-based model for PRS construction (VAE-PRS).
• Evaluation of VAE-PRS performance on biobank-level data for 16 blood cell traits.
• Utilizing Shapley additive explanations (SHAP) for model interpretability.

Main Results:

• VAE-PRS outperformed state-of-the-art methods in 14 out of 16 blood cell traits.
• The model demonstrated computational efficiency and robustness across different variant sets.
• VAE-PRS effectively captured interaction effects in high-dimensional genomic data.
• SHAP analysis provided insights into trait-associated genetic variants.

Conclusions:

• VAE-PRS offers a powerful deep learning-based approach for genetic risk prediction of blood cell traits.
• The model's ability to capture interactions and its interpretability advance personalized medicine.
• VAE-PRS facilitates genetic research by identifying novel trait-associated genetic variants.