Figure 1.

Overview of DeepBiomarker. (A) Data sampling process from EMRs to create case and control cohorts; data augmentation was applied to oversample AD + P patients to create balanced datasets. (B) Data embedding is a process of representing higher dimensional data in a lower-dimensional space while preserving the relevant properties of the original data. (C) Prediction by neural network with LSTM as the basic prediction unit. Perturbation-based contribution analysis was used to identify important features.