Google search algorithm helps track spread of cancer cells

Scientists are using the equations Google employs to predict the Web pages users visit to track the spread of cancer cells throughout the body.

“Each of the sites where a spreading, or ‘metastatic,’ tumour could show up are analogous to Web pages,” said Paul Newton, a mathematician at the University of Southern California, who has been working with cancer specialists at the Scripps Research Institute.

Google ranks Web pages by the likelihood that an individual would end up visiting each one randomly. These predictions are based on the trends of millions of users across the Web. Google uses something called the “steady state distribution” to calculate the probability of someone visiting a page.

“You have millions of people wandering the Web, [and] Google would like to know what proportion are visiting any given Web page at a given time. It occurred to me that steady state distribution is equivalent to the metastatic tumour distribution that shows up in the autopsy datasets,” Newton explained.

The dataset he’s referring to contains information about autopsy patients from the 1920’s to the 1940’s, who died before chemotherapy was available. By focusing on this group of patients, the researchers could track the natural progression of cancer, specifically lung cancer, without different treatments interfering with the data.

Out of fifty metastasis sites described in the autopsy reports, the scientists found that twenty-seven contained cancer that appeared to have spread from the lungs. Furthermore, just like with an individual browsing the Web, cells that break off from the original lung tumour and entered the bloodstream had a certain probability of progressing to different locations.

So, following Google’s example with search results, the researchers split the sites where the lung cancer spread to into two groups: first order and second order. In first order sites, tumour cells would most likely reach them by travelling directly from the lung. Tumours are more likely to reach second order sites by colonizing a first order site and then spreading to the second order location.

Using this approach, the researchers were able to estimate the average times it takes the cancer to spread to different parts of the body. The lymph nodes were the quickest to be affected by metastasizing lung cancer cells, with the adrenal gland and liver following close behind.

At the other end of the spectrum, lung cancer cells take so long to spread to the bladder and uterus that an individual with lung cancer would probably have died before those sites can be affected.

Unfortunately, the researchers’ dataset didn’t include records of the times when doctors noticed each new tumour. But the researchers could see how many tumours existed in each new site, input that information into the model, and calculate each progression.

“What we’re trying to do now is use this baseline model and make it patient specific, or at least subgroup specific to make more targeted predictions,” Newton said.

The researchers will also play with the model, searching for novel ways of reducing a cancer’s likeliness of spreading, for example, by isolating a key site in the body that would spread to other locations.