Fig. 1

Schematic overview highlighting which dataset cohorts are used in each stage of the modelling process. The left panel shows the training of the entire network using the UK Biobank healthy control cohort. The middle panel shows fine-tuning of the model weights using the EPAD super healthy fine-tuning control cohort in a transfer learning approach. The right panel shows the EPAD super healthy holdout and at-risk cohorts being passed through the network and the resultant latent vectors and data reconstructions being used to calculate individual-level deviations