Extended Data Fig. 1. Plate notation of ATM generative model.

M is the number of subjects; N_s is the number of records within s^th subject. All plates (circles) are variables in the generative process, where the plates with shade w is the observed variable and plates without shade are unobserved variables to be inferred; β is the topic weight; z is diagnosis-specific topic probability; t is the age at diagnosis for each diagnosis; β is the topic loadings, which are functions of age t; α is the (non-informative) hyperparameter of the prior distribution of θ. The generative process is described in the Methods and Supplementary Note.