Googling cancer: search algorithms can scan disease for patient risk

May 18, 2012


Network showing genes that are regulated by FOS and SP1 genes. It contains many literature-associated and highly correlated genes. Genes reported in the literature associated with pancreatic cancer survival are represented with larger circles. The absolute correlation coefficient of gene expression with survival in the screening dataset is shown in red. (Credit: C. Winter et al./PLoS)

The algorithm Google uses to rank search results can now scan cancers to see which molecules best reveal the risks patients face, researchers have found, Txchnologist reports.

By seeing how proteins are linked in a kind of molecular Facebook, search engine algorithms could also help unearth new targets for drugs to help combat tumors, investigators added.


The algorithm Google uses to rank which results pop up first in search queries, PageRank, orders results based on how other web pages are connected to them via hyperlinks.

Researchers modified PageRank to develop NetRank, which scans how genes and proteins in a cell are similarly connected through a network of interactions with their neighbors — “‘friends’ in the social network analogy,” said researcher Christof Winter, a medical doctor and computational biologist at Lund University in Sweden.

The investigators focused on pancreatic cancer, the most common form of which, pancreatic ductal adenocarcinoma, accounts for approximately 130,000 deaths each year in Europe and the United States. Very few tests exist to find out a prognosis for the disease — how it might progress, whether a patient might live or die.

The researchers used NetRank on about 20,000 proteins to see which ones were the best indicators for survival. They identified seven proteins that could help assess how aggressive a patient’s tumor is and guide clinicians to decide if the prognosis was worth trying chemotherapy or not.

As to how accurate prognoses based on these seven markers were, roughly speaking, “our markers are right in two-thirds of cases, and wrong in one-third,” Winter said. These markers were 6 to 9 percent more accurate at prognoses compared with those relying on conventional clinical parameters.


Monte Carlo cross-validation workflow to evaluate the accuracy of methods for ranking genes for outcome prediction (credit: C. Winter et al./PLoS)

In addition to improving prognoses of cancer, this research could also help identify new targets to help destroy tumors.

Currently Winter and his colleagues are analyzing DNA and RNA data from breast cancer. The hope is “to develop a DNA-based prognostic blood test for breast cancer patients,” he said.

Ref.: Christof Winter et al., Google Goes Cancer: Improving Outcome Prediction for Cancer Patients by Network-Based Ranking of Marker Genes, PLoS Computational Biology, 2012, DOI: 10.1371/journal.pcbi.1002511 (open access)