AI model can help determine where a patient’s cancer arose

Researchers from MIT and Dana-Farber Cancer Institute have developed a machine learning model called OncoNPC that may help doctors identify the origin of cancer tumors in the body. For a small percentage of cancer patients, doctors are unable to determine where their cancer originated, making it difficult to choose appropriate treatments. The OncoNPC model analyzes the genetic sequences of approximately 400 genes and accurately classifies at least 40 percent of tumors of unknown origin with high confidence, according to a dataset of around 900 patients. This improved classification allows for a 2.2-fold increase in the number of patients eligible for targeted treatments based on the specific origin of their cancer.

The inability to determine the origin of cancer tumors often prevents doctors from prescribing precision drugs that are typically approved for specific cancer types. These precision treatments are more effective and have fewer side effects compared to broad-spectrum treatments commonly used for cancers of unknown primary (CUP). By using the OncoNPC model, doctors can guide personalized treatment decisions for patients with cancers of unknown primary origin, potentially improving treatment outcomes.

The researchers trained their machine-learning model using genetic data from almost 30,000 patients diagnosed with one of 22 known cancer types. Subsequently, they tested OncoNPC on approximately 7,000 tumors with known origins, achieving an accuracy of about 80 percent. The model’s predictions were compared with germline mutations in a subset of tumors, which confirmed that the predictions were more likely to match the type of cancer most strongly predicted by the germline mutations. Additionally, the model’s predictions correlated with the expected prognosis for each predicted cancer type, further validating its accuracy and potential clinical utility.