This algorithm has been designed to spot these genetic indicators across many dog breeds with different forms of cancer: say, a bloodhound with lymphoma or a golden retriever with hemangiosarcoma. When the PetDx team began their study, they used blood from 224 of the dogs—ones known to either have cancer or not—to refine their algorithm. This “training set” helped PetDx determine a threshold for each genomic variation, defining what they should call signal rather than just noise.
Then they unleashed the algorithm onto data from the 876 other dogs. For each, it would render a binary answer: yes cancer, or no cancer. (For most cancers, it wouldn’t identify which form
→ Continue reading at Wired - Science