I train dogs to find lost people and human remains. There have been some case/control work with dogs who are given one petri dish with a type of cancer (bladder cancer if I recall) and five dishes with placebo tissue samples (i.e., no cancer cells present). The dog is trained to respond to the one with known cancer cells. The trainer changes up the placement of the known sample to ensure the dog is not hitting on the location rather than the scent.
There was an instance when the dog hit on a new sample of what was believed to be non-cancerous cells and rather than correct the dog, they rechecked the patient source of the sample to find that the dog was correct.
As my dog will find people underwater (because he has been trained on adipocere and decomposing human tissue underwater) as well as on land (training on all types of tissue and bone), I can’t tell you what part of the sample is stimulating him to respond (my nose doesn’t work as well as his).
So it is impossible to guess what element of the sample is causing the dogs to react, but there seems to be consistent reactions to the indicator sample using the same training methods we use for human remains detection.