The protein-interaction map is then overlaid with a gene-expression profile from a breast-cancer patient’s biopsy. Instead of looking at whether a patient with metastatic breast cancer is making more or less of one protein than a patient with a less aggressive form of the disease, the San Diego researchers were able to highlight protein pairs whose activity changed. They then looked for clusters in the interaction map where the activity level of a group of connected proteins was different in patients whose cancer eventually metastasized than in patients whose cancer did not. “Once you find the hot spots, you extract them from the hairball, and you have networks that correlate with metastasis,” says Ideker.
Ideker’s group discovered changes in the average activity of networks associated with the hallmarks of cancer, including metabolism, cell growth and division, and cell mobility. The researchers found that these changes in the activity of networks were better at predicting risk of metastasis than was analysis of gene-expression profiles alone. Looking at networks of proteins, “you see key changes you can’t see looking at individual genes,” says Collins. “Proteins rarely act individually.”
Ideker says that he is currently in discussion with several companies about how to develop the network approach into a commercial test. However, he cautions that because his group tested its approach on preexisting gene-expression databases, it needs further testing in breast-cancer patients.
Ideker says that his group is already applying the protein network to other diseases, with promising early results. Collins agrees that the approach is generalizable, and he says that it may allow for early detection of diseases besides cancer.