When someone gets cancer, many scientists believe, it begins with a mutation in what's called a cancer driver gene. That gene, they believe, causes the cell to replicate rampantly and create a tumor.
Researchers around the world are trying to identify these driver genes so lifesaving drugs that target cancer at the genetic level can be developed. But there's no widely accepted standard for identifying the genes, and many scientists are using different methods.
A group at the Johns Hopkins University is trying to solve that problem.
They have developed a software program using bioinformatics — the collection and analysis of biological data — to help identify cancer driver genes, and established a framework for evaluating the effectiveness of other methods out there.
The Hopkins group published a paper describing its research — "Evaluating the evaluation of cancer driver genes" — in the Proceedings of the National Academy of Sciences in October.
"The real challenge is to find the drivers," said co-author Rachel Karchin, an associate professor of biomedical engineering and oncology at Johns Hopkins. "It's like the needle in a haystack problem.
"For the past 10 years people have been developing statistical methods to do this, but it's really the Wild West. To identify which of these methods are best, there really are no rigorous standards, and that's what our paper attempts to do."
Scientists have identified 30 to 40 genes they believe are likely cancer drivers.
One, for instance, is believed to cause chronic myelogenous leukemia, an uncommon cancer of the blood cells that typically affects older people. Researchers developed a drug, Gleevec, that targets that gene.
Scientists believe there are more cancer drivers that have not been discovered yet, but identifying them is a challenge.
Each cell has about 25,000 genes. Further complicating things are so-called "passenger" mutations, which follow the cancer drivers, but can be difficult to distinguish from them.
20-20 Plus, the bioinformatics software developed by the Hopkins group, is aimed at making the job easier. It analyzes data from many tumors, and then uses an algorithm to predict likely cancer drivers.
John Quackenbush, a professor of computational biology and bioinformatics at Harvard University, said looking at cancer at the genetic level was like looking at a train barreling out of control.
Inside the train, he said, several switches might all be on, and it might not be clear which switch was causing the train to go.
"If we looked into thousands of trains and always the same switch was turned on, we could say, 'Aha, that's the one making the one driving the train forward,'" Quackenbush said.
"The problem with cancer is that it's not just one thing that drives it forward. What we're trying to do is understand in all these different trains what the right switches are. In our cancer cells there are 25,000 potential switches. We have a lot of possibilities, and not a lot of data to constrain it."
He said the framework developed by the Hopkins team will help cancer researchers make better predictions.
Otis Brawley, chief medical officer for the American Cancer Society, described gene research and immunotherapy as "hot" fields of cancer research. He said many researchers are developing huge databases of genomes to study cancer's origin.
He praised the Hopkins research.
"It's not uncommon now if you go to any of the cancer centers, they will want to take your tumor and sequence it to see what the mutations are," Brawley said. "The more tumors that we sequence, the more tumors we're able to gather information from.
"With a large computer to help us understand it, the more likely we are to help everybody."
In their paper, the Hopkins researchers compared eight methods of finding cancer drivers. Seven were developed by other researchers; the eighth was 20-20 Plus.
There is no gold standard for identifying cancer drivers, Karchin said, but there is a "bronze standard."
In testing the methods, it helped that scientists had already identified the 30 to 40 well-known cancer driver genes.
"We have the initial part, the sort of famous genes, so you can at least get a sense of are you getting those right as a first step," said lead author Collin J. Tokheim, a doctoral student at Hopkins' Institute of Computational Medicine and the Department of Biomedical Engineering.
If a method predicts a few genes are cancer drivers and none of them are in the list of well-known cancer driver genes, "there's something wrong with the method," Karchin said.
The researchers then compared the eight methods to determine if they overlapped with each other and found similar potential driver genes.
"If a method predicts 500 driver genes and only 10 of them were predicted by any other method, you might get suspicious," Karchin said. "A good method would make predictions that would be unique, but a majority of predictions would not be unique."
The team has shared its software online for others interested in cancer driver genes. Tokheim has made himself available to scientists who have questions about the research.
Karchin said the team's big takeaway was that the algorithms used to find cancer drivers could be improved.
"Many of them have a lot of false positives," she said.
"The real hope is to find drug targets with this work," she added.
"For most cancers, targeting a single gene is not enough. The hope is to find multiple drivers with each cancer type and develop combination therapies."