>In my mind, the variations on polygenic risks scores have convinced me that we need to build large-scale (whole-genome) models of disease that use nonlinear functions trained on extremely large-scale datasets (like UKBB) to build up wholistic predictive models that do a better job of encapsulating the complexity of biology and its relation to disease, to the point where we can actually start making useful treatments and cures for a wide-range of genetically determined diseases.

This is exactly where AI is going in the biomedical space. However, there's more than just the genome, you need to integrate multi-omics and some of the necessary tech hasn't been invented yet.