Fig 1. Pipeline for estimating networks, factor loadings, and predictive model for biological brain age.

Inferred factors $W\in {\mathbb{R}}^{p\times k}$ describe networks which are reproducible across the entire population, the subject-specific factor loadings $g_l^{\left(i\right)}$ are then used to predict brain age. Once the factor loadings are estimated as above, using one experimental data-set (we use CamCAN data in our experiments), we can then assess how these factors perform for brain age prediction on completely held-out data-sets; we demonstrate how the model generalizes well using HCP and ATR Wide-Age-Range datasets.