SAN FRANCISCO (GenomeWeb) – Using gene expression data from pediatric tumor samples, researchers at the California Kids Cancer Comparison (CKCC) project have been able to match 100 percent of their samples to possible treatments for the patients, according to Project Leader and University of California, Santa Cruz Biomolecular Engineering Professor David Haussler.
The project was first funded by the California Initiative to Advance Precision Medicine in August 2015 to build on the UCSC Treehouse Childhood Cancer Project, which was established to enable the sharing of genomic data from various kinds of pediatric cancer from hospitals and cancer centers.
At the Molecular Medicine Tri-Conference in San Francisco today, Haussler gave an update on the CKCC's progress, noting that its approach of primarily using gene expression data to match patients whose cancers come back, who have proven resistant to treatment, or who have little or no treatment options to therapies that may help them live longer or better lives, has been successful even beyond the initial goals of the project itself.
Out of 146 samples that passed CKCC quality control measures, the project's researchers found treatments they could present to clinicians for all of the samples, significantly exceeding the initial goal of finding treatment leads for 20 percent of the samples, Haussler said. Further, he added, 79 percent of the treatment leads were found by an automated analysis workflow developed by CKCC researchers, while only 21 percent of the matches had to be determined manually by CKCC data analysts.
CKCC has also built a gene expression map for many types of pediatric cancers, which contains about 11,340 vectors of RNA expression, Haussler said. The gene expression analysis workflow is done in two parts. In the first step, pediatric tumors are molecularly classified based on an analysis of their gene expression. And in the second step, targeted therapies are identified as treatments that could be considered for a given patient, based on the increased expression of druggable gene targets in the patients' tumors.
"We now know so much about the genome that we can go beyond standard of care in these cases," Haussler said.
He also detailed two patient cases where the CKCC workflow had matched children with treatments for their advanced cancers. In both cases, the children passed away, but they both gained years of life that they otherwise would not have had, he said.
There is a big problem, however. The CKCC approach utilizes big data/pan-cancer information to derive clinically meaningful relative measurements of gene expression, according to Haussler. In order to derive relative measurements, researchers (or artificial intelligence systems) require access to lots of carefully processed data. But getting the data is not easy — most genome data is held in silos and not standardized for exchange, and no one institute has enough data on its own to make enough progress on cancer research, Haussler said.
What is needed to end this problem is a network for sharing data. This is the network that the Global Alliance for Genomics and Health (GA4GH) is now trying to build, he added, noting that 500 institutional members from 70 countries and more than 150 companies are now taking part, building their ethics and standards, and deciding on how their network will actually work in order to facilitate data sharing around the globe.
In October 2017, GA4GH unveiled its five-year strategic plan, calling on its members to develop new data-sharing standards for use in major international genomic data initiatives.
Until the data-sharing problem is solved, there are several issues that could continue to affect CKCC's work, Haussler noted.
"Scale is critical here — the first several times we went through this workflow, it was very human-intensive, but we're automating more and more," he said. "But it's not without its glitches, and there's a lot more to be automated, like literature search."
However, he added, there's one reason computers aren't better at certain things like literature searches than human beings, and that's because the training sets they've been subjected to haven't been large enough, the result of not enough data being shared.
Also, he said, "We want to anticipate the cancer's next move. We find that it's like whack-a-mole."
One thing he and his team would like to do is determine how the cancer will adapt and maybe apply a combination therapy in order to head that off. CKCC's gene expression workflow approach does have that potential, Haussler noted, but only if it gets smart enough to offer clinicians options that take possible tumor adaptations, as well as possible toxicities, into account, as combination therapy toxicities are a particular concern in pediatric oncology.
"I envision a future where we think about this like a computer game," where smart people and computers play this game against cancer, Haussler said, adding, "I envision us winning this game in the future," if we start sharing information.
He also noted that his team would eventually like to take full advantage of recent advances in liquid biopsy technology. They are working with people that are developing liquid biopsy technology to capture data from exosome and circulating tumor DNA data, Haussler said, and as this technology gets more accurate, the CKCC team would like to be able to measure everything it needs from each patient from blood samples rather than from biopsies of solid tumors.