October 18, 2010

Computer program helps researchers predict pancreatic cancer

Using a computer program, researchers from Johns Hopkins have predicted which changes in the DNA code may cause pancreatic cells to become cancerous and deadly. The investigators say the findings could lead to more focused studies on better ways to treat the disease, which has only a 5 percent survival rate five years after diagnosis. In a report on the work published in the Sept. 15 issue of Cancer Biology and Therapy, the investigators note that cells in the body accumulate DNA changes over time as a result of the normal aging process and from different toxins in the environment. A cancer cell may have acquired thousands of different DNA changes, but in actuality only a few of those changes may lead to cancer. Determining which of these changes cause the cell to be cancerous is challenging.

“A list of thousands of these changes detected by sequencing the DNA from a group of patients is daunting. Where do you even start?” asks Hannah Carter, a graduate student of Rachel Karchin’s, of the Department of Biomedical Engineering and the Institute for Computational Medicine and lead author of the study.

One approach to finding cancer-causing genes has been to look for those with many more DNA changes than expected. But because some culprit genes don’t contain many changes, a computer program was designed to detect important changes, regardless of whether the change was found in one or many patients.

Researchers at the Sol Goldman Pancreatic Research Center at Johns Hopkins sequenced the DNA in pancreatic tumors from 24 patients and compared them to DNA sequences from healthy tissue from the same patients. The comparison turned up many kinds of DNA changes. However, there were 963 DNA changes of a specific type where one letter of the genetic code is modified that was unique to the pancreatic cancer cells that the group chose to focus on. The sequence data were then given to Karchin’s team for assessment.

“Our role in all this is to try to distinguish which of these mutations are worth investing lab resources, money and time in following up,” Carter says.

The researchers designed the computer program by listing all individual genetic changes suspected of causing cancer and those highly unlikely to cause cancer. The program then used 70 different predictive features for each DNA change, such as the DNA sequence and structure of the resulting protein, to identify any of the distinguishing characteristics of driver mutations—those DNA changes that contribute to cancer—compared with other genetic changes.

Once the researchers had a computer program that could successfully distinguish between DNA changes that are cancer-causing and those that are not, they used it to assess the 963 changes found in the pancreatic cancer cells. The program gave each of the 963 DNA changes a score between zero and one, with zero meaning a likely driver and one meaning not likely to be a driver mutation.

“Our results can help cancer biologists set up experiments to see how important these DNA changes really are in pancreatic cancer and whether or not they are good drug targets for potential treatments,” Karchin says.

Additional authors of the manuscript are Josue Samayoa, of the Johns Hopkins Whiting School of Engineering, and Ralph H. Hruban, of the Johns Hopkins School of Medicine.

This research was supported by funds from the National Science Foundation, the National Cancer Institute and a Department of Defense graduate fellowship.