Extended Data Fig. 9: Neural network architecture of the AI-based model.

a, The neural network takes as input a night of nocturnal breathing. The main task of PD prediction consists of a breathing encoder, a PD encoder, a PD classifier and a PD severity predictor. We also introduce an auxiliary task of predicting the subject’s qEEG during sleep. We include also two discriminators for domain-invariant transfer learning. b, The detailed architecture of each neural network module in the model.