Google PageRank algorithm, Markov chains, and cancer.

# Fighting Cancer With the Google PageRank Algorithm. Sort Of.

The Citizen's Guide to the Future
March 27 2013 12:02 PM

# Fighting Cancer With the Google PageRank Algorithm. Sort Of.

In cancer parlance, metastasize is a four-letter word. Metastasis is when cancer cells break off of the primary tumor to surf the bloodstream and set up shop in new organs and body areas. Thankfully, researchers are developing a new way to combat this life-and-death variety of hide-and-go-seek, a strategy borrowed from one of the best dang seekers around—the Google PageRank algorithm.

To start, researchers gathered data from the autopsy reports of lung cancer patients from 1914-1943. The time period is important because it predates interventions like radiation and chemotherapy, which allowed the study to look at cancer left to its own jerky devices. The researchers could then load the many real-life cancer trajectories into “an algorithm similar to Google PageRank … to gain important insights about the spread patterns of lung cancer,” as a press release put it. The findings were published recently in Cancer Research.

“Basically, we are doing the inverse of what Google does,” says Paul Newton, an applied mathematics specialist at USC Viterbi School of Engineering. “They know the transition probabilities and compute the steady-state, we know the steady-state and compute the transition probabilities.”

In other words, Google knows where you could go and uses a mathematical system called Markov chains to determine how likely you are to get there. The researchers knew where cancer did go and used similar equations to investigate how it got there. (It won’t surprise anyone in the math crowd that Google PageRank is based on these Markov chains, or sequences of random variables used to determine the probability of going from one “state” to another. But the analogy is a deft move by the Scripps PR department as most of our ears do not exactly prick at the mention of algorithms, statistical models, or transition probabilities.)

Newton and his team have also used this sophisticated system of mathematical equations to identify certain regions of the body that work as “sponges” and “spreaders.” For instance, lung cancer cells often travel to sponge sites like the liver, lymph nodes, and bones, but generally don’t branch out further from these posts. Conversely, the kidneys and adrenal gland seem to be spreaders, veritable launch pads for later stages of invasion.

This research calls into question the prevailing theory that metastatic lung cancer marches in one direction. Even more interesting, by identifying sponges and spreaders, the researchers may have given us a new way to target treatments and lessen or prevent the spread of cancer. If we could learn more biologically about what makes a sponge a sponge, perhaps we could alter these parts of the body to perform that function even better. Newton described this to me colorfully as a “Hotel California,” a place where “a circulating cancer cell enters but never leaves.”