With the help of machine learning, scientists have identified a plethora of previously-unidentified drug targets for breast cancer, cervical cancer, glioblastoma and more.
In a study published Jan. 11 in Cancer Cell, scientists from the Wellcome Sanger Institute reported the identification of 370 drug targets across 27 different types of cancer. The results are useful not only for coming up with new precision drug targets, but also for building out the Cancer Dependency Map, a joint project between Wellcome Sanger and the Broad Institute that seeks to inform strategies for personalized cancer treatments.
“This work exploits the latest in genomics and computational biology to understand how we can best target cancer cells,” study co-lead Mathew Garnett, Ph.D. said in a press release. “This will help drug developers focus their efforts on the highest value targets to bring new medicines to patients more quickly.”
The long list of candidates were the result of an analysis on a large set of data from the Cancer Dependency Map, which was created by performing CRISPR-Cas9 alterations to every gene inside 930 different human cancer cell lines, all at the same time—a total of nearly 18,000 genes. The researchers used machine learning to delve into the dataset and figure out which genes, proteins and pathways were critical to cancer cell survival.
After narrowing down the candidates, the researchers linked them to clinical biomarkers, which would allow them to see what patients might benefit from treatments aimed at various genetic targets. “Notably, almost all targets had an associated marker, which is important as genetic evidence between a target-disease indication increases 2- to 4-fold the probability of FDA approval of a drug during clinical development,” the researchers wrote in their paper.
Additional analyses suggested that at least 30% of all cancer patients carry a biomarker for one of the identified targets, roughly double the 14% of patients who are thought to be candidates for therapies that target the cancer genome, the team wrote.
However, in many cases drugs will never be developed for those targets, they added, and the presence of genomically-linked biomarkers varies by cancer type, with some types having little or no change in new targets based on the findings.
“This emphasizes that novel approaches are still needed to enable expansion of the current repertoire of new medicines for patients,” the researchers said. Their results also suggest that multiomics biomarker tests—that is, tests that look at multiple levels of biological information together, such as assessing both gene and protein expression—could help expand therapy targets in patients.