An AI model developed by researchers at Massachusetts Institute of Technology (MIT) and Dana-Farber Cancer Institute could potentially make it easier to target treatments for patients with cancer of unknown primary origin, following a study in which the model was able to “accurately classify at least 40 percent of tumours of unknown origin with high confidence”.
Researchers used machine learning to create a computational model called OncoNPC, capable of analysing approximately 400 genes that are often mutated in cancer. The model was trained on data from nearly 30,000 patients who had been diagnosed with one of 22 known cancer types, before it was tested on around 7,000 tumours that the model had not seen before, where the site of origin was known by researchers in advance.
They found that OncoNPC was “able to predict their origins with about 80 percent accuracy”, and for tumours with high-confidence predictions, “its accuracy rose to roughly 95 percent”.
The model was then put to work analysing a set of 900 tumours classified as cancers of unknown primary origin, with high-confidence predictions for 40 percent of the tumours.
Researchers went on to compare the predictions made by the model with an analysis of inherited (germline) mutations in a subset of tumours, capable of revealing whether patients have a genetic predisposition to develop particular cancers. They found that “the model’s predictions were much more likely to match the type of cancer most strongly predicted by the germline mutations than any other type of cancer.”
Alexander Gusev, associate professor of medicine at Harvard Medical School and Dana-Farber Cancer Institute and senior author of the study, explains how “a sizeable number of individuals develop these cancers of unknown primary every year, and because most therapies are approved in a site-specific way, where you have to know the primary site to deploy them, they have very limited treatment options.”
The research team has also used the model to identify an additional 15 percent of patients “who could have received an existing targeted treatment, if their cancer type had been known” rather than receiving more general chemotherapy drugs. Professor Gusev adds: “That potentially makes these findings more clinically actionable because we’re not requiring a new drug to be approved. What we’re saying is that this population can now be eligible for precision treatments that already exist.”
In other news around AI, last month we covered the University of Aberdeen’s work with NHS Grampian and Kheiron Medical Technologies on an AI breast screening technology for breast cancer, which can potentially detect abnormalities traditional screening methods would have missed. We also recently looked at the use of AI to design a bespoke nanoparticle to deliver a drug molecule, mRNA, to cancer cells.
In addition, we covered the DHSC’s announcement of £21 million in funding for AI to “help diagnose patients more quickly for conditions such as cancers, strokes and heart conditions”.
