An open-source machine learning algorithm can predict an individual’s response to a particular cancer medication with more than 80% accuracy.
Precision or personalized cancer treatment relies not on clinical pathways or protocol, but on doctors being able to choose a drug that targets the individual molecular profile of a patient’s tumor. However, the genetic markers underlying many cancers are not fully understood.
The answers lie in the cancer-related data that is gathering at a tremendous rate worldwide. Searching this information for correlations is a task suited to a branch of artificial intelligence called machine learning.
Researchers from the Georgia Institute of Technology recently introduced an open source support vector machine (SVM)-based algorithm. It uses the gene expression profiles of cancer cells to predict the response of individual cancers to chemotherapy drugs.
The method was accurate when tested on cancer cell models in the lab, and this latest study looked at accuracy in human malignancies.
The SVM-based algorithm analyzed data from 152 patient records.
To “train” the system, details on 114 records, including the RNA sequence of tumors, treatments used and the outcome. The teams then used 38 records to test the system’s ability to use the RNA sequence to predict which drugs were most likely to reduce the size of the tumor.
Data from ovarian, lung, breast, liver and pancreatic cancers, which all used the same medications, were examined. Researchers explained the model’s predictions were based on the drug used, regardless of cancer type.
The system used the data to produce a chart that compared the likelihood of each drug having an effect on the patient’s specific cancer. It predicted the medicine that provided the best outcome in 80% of cases.
Researchers believe the system could help doctors choose the chemotherapy drug that is most likely to shrink the tumor in any individual patient. It has particular use, they argue, in guiding decisions for those who have failed first-line treatment and are faced with multiple options.
“By looking at RNA expression in tumors, we believe we can predict with high accuracy which patients are likely to respond to a particular drug,” said John McDonald, a Professor in the Georgia Tech School of Biological Sciences and Director of its Integrated Cancer Research Center.
“This information could be used, along with other factors, to support the decisions clinicians must make regarding chemotherapy treatment.”
Cancer centers and hospitals are being encouraged to use the technology, which is available as open source software.
Machine learning becomes more accurate as more data is entered, and even better predictions could be achieved if data such as family history and demographics were included, the team believe.
Huang C, Clayton EA, Matyunina LV, McDonald LD, Benigno BB, Vannberg F, McDonald JF. Machine learning predicts individual cancer patient responses to therapeutic drugs with high accuracy. Sci Rep 2018; 8:16444.
Georgia Institute of Technology. (2018) Open source machine learning tool could help choose cancer drugs [Press release]. November 6, 2018. Available from: https://www.eurekalert.org/pub_releases/2018-11/giot-osm110618.php (accessed February, 2019).