DREAMing Big to Solve Cancer Research Challenges
Earlier today, at the DREAM7 Conference in San Francisco, representatives from the two best-performing teams in the NCI-DREAM Drug Sensitivity Prediction Challenge took to the stage to unveil their methods. Members of the two teams, from Helsinki, Finland, and Dallas, TX, along with 50 other teams from around the world, spent several months this year putting their heads together to work on two related scientific challenges.
NCI-DREAM is part of a larger project known as DREAM (Dialogue for Reverse Engineering Assessments and Methods). Now in its seventh year, DREAM is one of a growing number of biomedical research endeavors that take advantage of crowdsourcing—outsourcing a task to a group or community through an open call or contest.
NCI partnered with DREAM to support innovative approaches to cancer research and to capitalize on the breadth of knowledge of the research community to develop better treatments for cancer patients.
Crowdsourcing "takes advantage of the strength of the broader community to expand our understanding of some difficult scientific problems," noted Dr. Dan Gallahan, deputy director of NCI's Division of Cancer Biology and one of the organizers of the NCI-DREAM challenge.
"The project was born to try to help the research community understand the limitations and strengths of our own methods," explained Dr. Gustavo Stolovitzky, an IBM computational biologist who has been a driving force behind the project. "Crowdsourcing also allows us to determine what the best method is for the solution that we are seeking."
Participants in the two-part challenge were asked to use a vast array of genomic information to build computer models that can predict the sensitivity of cancer cell lines to a set of small-molecule compounds or combinations of compounds. The goal of sub-challenge 1 was to predict the sensitivity of 18 breast cancer cell lines to 31 previously untested compounds, while the goal of sub-challenge 2 was to predict the activity of pairs of compounds on a diffuse large B-cell lymphoma (DLBCL) cell line.
Genomic data for NCI-DREAM were provided by Dr. Joe Gray of Oregon Health and Science University and Dr. Andrea Califano of Columbia University.
"Drug sensitivity prediction from genomic profiles is the core problem of personalized cancer medicine…. At the same time, the task is fascinating due to the challenges it poses [for] computational analysis," wrote Dr. Elisabeth Georgii, a postdoctoral researcher at Helsinki Institute for Information Technology HIIT, Aalto University, in an e-mail message. Dr. Georgii is representing TeamFIN, the best-performing team in sub-challenge 1.
The best-performing team in sub-challenge 2 hails from the University of Texas Southwestern (UTSW) Medical Center. Dr. Yang Xie, an assistant professor in the Department of Clinical Sciences at UTSW, is representing her team at the meeting. (A complete list of members from the best-performing DREAM challenge teams is available online .)
The More the Merrier
In addition to a speaking invitation and travel expenses to the DREAM7 Conference in San Francisco for a team representative, the best-performing teams will publish a peer-reviewed Nature Biotechnology paper on the sub-challenge in which their method was the best performer.
If more people participate, it's more likely that we will find a method that will really hit the nail on the head.
—Dr. Gustavo Stolovitzky
"Incentives can go a long way to gather more people who try to solve the challenge. And if more people participate, it's more likely that we will find a method that will really hit the nail on the head," Dr. Stolovitzky noted.
The long-term goal of NCI-DREAM is to apply what was learned from the challenge to improve treatments for cancer patients. "This is the start of being able to make predictions of how to treat a patient based on his or her molecular profile," Dr. Gallahan explained. With that goal in mind, NCI plans to support the subsequent experimental validation and development of the top performing models in the challenge.
"Cancer is such a complex disease that we want to engage as many people in cancer research as possible" he continued. "And if crowdsourcing is one way that we can do that, then that's another tool in our arsenal."