More than a million people are diagnosed with prostate cancer each year, and in the U.S., approximately one in nine men will develop prostate cancer in their lifetime, according to the National Cancer Institute. Unfortunately, it tends to be problematic from a treatment perspective -- it's frequently non-aggressive, making it difficult to determine which, if any, procedures might be necessary.
Encouragingly, an international team of European researchers from the University of Copenhagen, the European Molecular Biology Laboratory in Heidelberg, and other leading institutions have made progress in developing a computer model that can differentiate between aggressive and non-aggressive prostate cancer cases based on certain genes. They describe their work -- which is currently being implemented in an unnamed clinic in Germany -- in the December 10 edition of the journal Cancer Cell.
"If we have a patient with a particular set of [genetic] mutations, we can use the model to predict the most likely next mutation that the patient will experience at some point -- and how it will affect the patient's clinical situation. As an illustration, we can predict with some probability that if you have mutation A, you are likely to get mutation B before you get C," group leader and study coauthor Joachim Weischenfeldt said in a statement. "We can also predict if the next mutation is likely to change the clinical outcome of the disease."
To train the algorithm, the team collected patient data from 300 men with early-onset prostate cancer -- prostate cancer affecting those under the age of 55 -- who'd had their entire cancer genome sequenced. (Tumors in their infancy are easier to analyze, the researchers said, and give a clearer picture of the first several mutations and alterations.)
In the course of training, the model identified a mutational mechanism involving APOBEC, an enzyme that the researchers hypothesize might trigger tumor growth. It also surfaced a novel oncogene -- a gene that can transform a cell into a tumor cell -- called ESRP1, which the researchers believe could be used as a new biomarker to detect whether a patient will have aggressive prostate cancer.
"We hypothesise that this enzyme mutates the prostate cells at a low but constant rate. Each time the cell divides, APOBEC is likely to cause mutations. If you have early-onset prostate cancer, you may have a couple of mutations caused by APOBEC. Twenty years later, you may have 10-20 mutations," coauthor Jan O. Korbel said.
The results come about a month after Google AI, the artificial intelligence (AI) research division at Google, detailed in a blog post a model that uses the Gleason score -- a grading system that classifies cancer cells based on how closely they resemble normal prostate glands -- to detect problematic masses in patient samples. In testing, Google AI's system achieved an overall accuracy of 70 percent, beating the 61 percent achieved by the U.S. board-certified pathologists who participated in the study.