Sorting through seemingly endless numbers of genetic mutations in order to pin down the few implicated in cancer can be a big headache for researchers. But a team at Johns Hopkins University is developing software that plays a sort of guessing game to focus on protein sequences called missense mutations that evidence minute variations from normal patterns.
Called CHASM — for Cancer-specific High-throughput Annotation of Somatic Mutations — the software employs a machine-learning technique to determine which cancer-causing mutations are drivers or passengers by parsing through numerous characteristics assigned to cancer-causing mutations using numerical values. "Up to just a few years ago the state of cancer biology was that there were just key genes that were important to cancer — oncogenes, tumor suppressors," says Rachel Karchin, an assistant professor of biomedical engineering at Johns Hopkins who supervised the effort. "What's changed in the field now is that it's not about a few genes, it's about thousands of genes. And this creates a problem that there are all these mutations. The belief is that only a small subset of these is actually going to be important in terms of giving the cancer cells a selective growth advantage, but how to tease them out is the motivation."
The developers used CHASM to sort through some 600 possible brain cancer-causing genes.
Karchin says that the biggest challenge in making this approach feasible is finding features to describe the mutations. "Essentially, you have to quantify them some way and hope that the numbers that you calculate can represent the properties of the mutations that make them driver mutations or passenger mutations," she says. "It takes quite a bit of creative thought and pooling of databases and resources trying to integrate a lot of tools [and] information out there to bring it together to get the most representation of these mutations as possible. … The hope is that the machine-learning method can discover patterns within those representations that help it distinguish the drivers from the passenger."
So far, CHASM is available for use as a Web-enabled application. Karchin and her team are encouraging users to apply their CHASM approach to identify top-priority mutations for other cancers.