Skip to main content
Premium Trial:

Request an Annual Quote

Newbie Watch: Lancet Ovarian Cancer Study Puts Correlogic Systems in the Limelight

Premium

A recent study of ovarian cancer in The Lancet [BioInform 02-11-02] heralded the advent of protein profiles as possible diagnostic tools. It also put the spotlight on a tiny bioinformatics company in Bethesda, Md., which provided the artificial intelligence-based algorithm that identified the biomarkers: Correlogic Systems.

About three years ago, “it literally began on the back of a napkin over lunch,” recalled Peter Levine, one of the company’s two founders and its president and CEO. That was when he met with researchers at the FDA and discussed the idea of looking for patterns in protein profiles from patients. Knowing that biochemist-turned-bioinformaticist Ben Hitt, Correlogic’s other cofounder and its chief scientist and technology officer, had some experience with pattern discovery, Levine approached him with the idea of forming a company based on turning biological patterns into diagnostics. “Ben and I then basically abandoned everything else that we were doing and formed Correlogic” in May of 2000, said Levine, who was president of an international trade and investment company at the time.

The company’s core technology is a patent-pending pattern discovery algorithm called Knowledge Discovery Engine. This iterative algorithm combines elements from genetic algorithms and self-organizing adaptive pattern recognition systems. It operates by randomly selecting small sets of candidates from a training data set and subjecting them to a fitness test, until the best set is found. In the ovarian cancer study, for example, it assessed the sets’ ability to segregate the training data into two clusters, women with or without cancer. The researchers used the beta version of a research application software based on the algorithm called Proteome Quest in the study. What is unique about the approach, Hitt said, is the “evolutionary component” as well as the fact that “it continually learns over time,” meaning that rather than sticking with the pattern it found in the training set, it can continue to incorporate new data, which may generate new patterns.

The computational requirements are minimal: “Proteome Quest runs on any garden variety of personal computer,” said Hitt. In fact, he tested it on new mass spec data for the ovarian cancer study on his notebook computer while on a plane to Japan. Although it is possible to find the “ideal” set of markers within an hour, he said, it took the researchers about a month to derive the model they reported in the Lancet study.

So far the algorithm has only been validated by peer review for finding patterns in protein expression data, but is not limited to this application. “You can use any biological data that would have some amplitude associated with it,” said Hitt, such as gene expression data. He could imagine applications in fields as diverse as drug toxicity studies, epidemiology, or process quality control.

At the moment however, Correlogic, which has grown to four employees and is located just a few miles from the NIH, does not have any partners or clients other than the NCI/FDA, with whom it is about to enter a longer-term research relationship. It has also filed a shared patent with the NCI/FDA on the general process of using patterns of molecular expression as a diagnostic.

But the Lancet study has certainly put Correlogic on the map, and the company is currently “culling through all the inquiries to find the best partners to be working with,” said Levine. Correlogic’s preferred business model is to enter research partnerships that will result in intellectual property in the form of diagnostic models. “The goal would certainly be to have a diagnostic test,” said Levine, but that road could be long. For other applications, the software may also be licensed in a more traditional way, he said.

Though the company is financed privately now, it has approached venture capital firms and possible partners in order to be able to expand. “There are a million different ways that we can approach this,” said Levine. “It’s a matter of coming up with the best partnership arrangement to carry out the goal of developing this portfolio of intellectual property.”

As far as he can see, there are no competitors on the horizon. But this might have changed with the recent high-profile publication: “We probably do now have a lot of competitors, but we are not sure exactly who they might be,” he said.


— JK

Filed under

The Scan

Fertility Fraud Found

Consumer genetic testing has uncovered cases of fertility fraud that are leading to lawsuits, according to USA Today.

Ties Between Vigorous Exercise, ALS in Genetically At-Risk People

Regular strenuous exercise could contribute to motor neuron disease development among those already at genetic risk, Sky News reports.

Test Warning

The Guardian writes that the US regulators have warned against using a rapid COVID-19 test that is a key part of mass testing in the UK.

Science Papers Examine Feedback Mechanism Affecting Xist, Continuous Health Monitoring for Precision Medicine

In Science this week: analysis of cis confinement of the X-inactive specific transcript, and more.