Genome sequencing and pharmacogenomics advances are already moving the diagnosis and treatment of cancer from an approach based roughly on a tumor's anatomical origin to one that identifies and targets specific genetic mutations. Now this new view of cancer is evolving further as better bioinformatics tools allow researchers to characterize the genomes of different cell populations in tumors, track how they evolve, and devise ways to use this information for prognosis and therapeutic decisions.
Bringing these methods to the clinic is a key priority for cancer computational biologists at the Broad Institute, who are focused on developing, using, and freely sharing better tools to tackle these challenges.
“It will be important to use these sensitive tools in clinical trials and, later on, in clinical practice to find mutations that could be actionable or at least affect your prognosis for the patient,” says Gaddy Getz, who directs the Cancer Genome Analysis Group at the Broad. Getz says the challenge is to be able to detect rare mutations occurring in very few cells in a sample using the data from standard exome and genome sequencing.
Publishing in Nature Biotechnology, Getz and a group led by Kristian Cibulskis showed that their tumor-mutation-detection tool, MuTect, might be just what the doctor ordered for identifying mutations in samples of tumor cell populations that are highly contaminated by non-tumor DNA, as well as in very small subpopulations of tumor cells that might resist initial drug treatments and present a future threat of relapse or metastasis.
The Broad team established benchmarking methods to compare various informatics approaches to detecting and verifying such rare mutations. Cibulskis, a lead developer of MuTect, explains that there's often a tradeoff between sensitivity and specificity. A supremely sensitive method would "call" a mutation based on any single bit of evidence, but would present a lot of false positives. “So, specificity and sensitivity are a balance,” he says.
The team devised a benchmarking mechanism "in which all methods were tuned to have the same specificity—the same error rate," Cibulskis explains. "We then measured the methods’ sensitivity and how it changed with the depth of sequencing and the allelic fractions—the fractions of DNA that contain the mutations.”
They found that the smaller the allelic fraction, the more sensitive MuTect was compared to other methods. “MuTect achieved a sensitivity two to three times greater than other methods … detecting mutations using only three to four reads,” says Getz.
Another of their benchmarking tools is “virtual tumor.” It measures the sensitivity versus specificity of mutation-calling tools. “We take a sample that was sequenced twice, and call one a tumor and the other normal,” Getz explains. Since they are comparing the same sample, “every mutation we detect is a false positive, because there shouldn’t be any difference between the samples.”
In addition to other tools they are making freely available to the research community is a tool called “ABSOLUTE,” developed by Scott Carter of Getz’ team. Introduced last May in Nature Biotechnology, it uses copy number alterations to quantify “what fraction of cancer cells has a certain mutation, and can distinguish clonal from subclonal mutations,” Carter says.
The researchers showed that marrying these two tools allows experimenters “to detect rare events and then quantify what fraction of cancer cells they belong to,” says Getz. They demonstrated its use in a study of the evolution of chronic lymphocytic leukemia cell populations in response to treatment, published last month in Cell.
Cibulskis emphasizes that, while sequencing costs have come down, the cost of computation is not keeping pace. The ability to assess the performance of algorithms in terms of sensitivity and specificity is critically important, he says, because “we are in the phase where we are still looking for the optimal method."
“We are still focusing on how well the tools are performing the tasks,” adds Getz. “But, once we reach a good enough performance—and I think we are now at that stage—the focus can shift to making it more efficient in terms of less computing resources needed to perform it.”
Making an analogy to the history of the automobile, Getz says, “We are still in the stages of building the first car … while working towards making the cheapest car.”
MuTect, ABSOLUTE, and numerous other bioinformatics tools are freely available for noncommercial use, and “we’re exploring license agreements for commercial organizations,” Cibulskis says.
Post a Comment