[possibilistic]

This is literally the field I'm preparing to go into after I get my PhD in systems bio. The absolute best way to develop models for markers is to study metabolomics data for a huge sample size.

Take, for example, urine samples from a large number of at-risk persons and see how a specific type of cancer manifests within the group. Run the samples on GCxGC-MS and do high-order statistical comparisons or train an ML model to pick up on the markers. Given the proper treatment, these may correlate to breakdown products attributable to the development of cancerous cells.

You can model any metabolome too: blood, breath, sweat, biopsy... Urine is probably the least invasive. You should learn a thing or two about how elimination works: glucuronidation and other reactions that modify substrates to be polar enough.