Genome-wide association study-based deep learning for survival prediction.

Area of Science:

• Genomics
• Biomedical Informatics
• Ophthalmology

Background:

• Accurate survival prediction is crucial for personalized medicine and disease management.
• Genome-wide association studies (GWAS) offer vast genetic data for developing predictive models.
• Deep learning shows promise in biomedical prediction but is underutilized for survival analysis with GWAS data.

Purpose of the Study:

• To develop and implement a multilayer deep neural network (DNN) survival model for accurate and interpretable survival predictions.
• To leverage rich GWAS data for enhanced prediction of disease progression.
• To apply the model to age-related macular degeneration (AMD) for personalized risk assessment.

Main Methods:

• Developed a multilayer deep neural network (DNN) survival model.
• Utilized GWAS data from over 7800 AMD patients from two large clinical trials.
• Compared DNN model performance against other machine learning-based survival models via simulation studies.

Main Results:

• The DNN survival model achieved high prediction accuracy (c-index = 0.76) in AMD patients.
• The model outperformed several existing survival prediction methods.
• Successfully identified clinically relevant risk subgroups by analyzing complex genetic variant structures.
• Provided subject-specific predictor importance for personalized insights.

Conclusions:

• The developed DNN survival model is effective for accurate and interpretable survival prediction using GWAS data.
• This approach offers valuable insights for personalized early prevention and clinical management of AMD.
• Highlights the potential of deep learning in integrating genetic data for complex disease prediction.