|
Abstract:
|
Recent advances in deep learning have made extraordinary achievements in establishing flexible and powerful prediction models. However, the application of deep learning in biomedical research is limited. The genome-wide association studies (GWAS) of Age-related Macular Degeneration (AMD), a progressive eye disease, is the first and most successful GWAS research, where the massive GWAS data provide unprecedented opportunities to study disease risk and progression. Motivated by the need to establish a flexible and reliable prediction model for AMD progression, we develop a novel framework, which builds deep neural networks on time-to-event outcomes to effectively extract features from the wealthy GWAS data. Specifically, we employ the Tensorflow framework to obtain high computational efficiency. Finally, using data from two large randomized clinical trials on AMD progression, Age-Related Eye Disease Study (AREDS) and AREDS2, we apply our method to develop, evaluate and validate a novel prediction model to predict the risk of progression to late-AMD given the patient’s age and genetic profile. The result provides valuable insights into early prevention and tailored intervention.
|
