Skip to main content
Premium Trial:

Request an Annual Quote

GNS to Use Computer Modeling to Classify Patient Populations in Cancer Clinical Trials


Gene Network Sciences is using its computational modeling technology in a partnership with the Mary Crowley Medical Research Center to classify patient subpopulations and identify biomarkers associated with response in cancer clinical trials.

The agreement, announced this week, marks the first time that the biosimulation firm's technology will be used in a clinical setting. GNS CEO Colin Hill told Pharmacogenomics Reporter that it will be the first "large-scale" project in which the company will use gene-expression data from patient tumor samples to "reverse engineer" molecular pathways that guide cancer drug response.

To date, GNS has primarily used its modeling platform in preclinical research collaborations with pharmaceutical firms, and has relied on data from human cell lines and mouse tissue to build computational models of biological networks. The company's core technology, called network inference, uses gene-expression data and other information as input to reconstruct biological networks that can serve as predictive models of response to perturbations. "We've modeled patient data before, but not in a big way," Hill said.

In the collaboration with the Crowley Center, Hill said that the "end goal" will be identifying biomarkers that separate responders from non-responders in a range of Phase I and Phase II clinical trials. Hill said that the partnership is not focused on any particular type of cancer or class of drug. The center will provide GNS with gene-expression data from the Affymetrix GeneChip platform from around 20 to 30 patients per year, and GNS will develop "causal models" of patient response for each trial.

Financial details of the agreement were not provided, but GNS said that it will be compensated on a "per-patient basis."

While the agreement nudges biosimulation into the realm of personalized medicine, Hill stressed that "the idea of an individual computational model for each patient would be extremely challenging," and perhaps not even feasible with current technology. Instead, he said, the company hopes to use its technology to break patient groups into subpopulations based on their response to certain compounds, and from there dig a bit deeper into the "black box of human biology" to identify specific mechanisms responsible for "linking a drug to a particular end point."

Hill is confident that computational modeling will have clear advantages over bioinformatics analysis of microarray or proteomics data, because it will generate "more accurate biomarkers" for classifying patient populations.

The project is not without its risks, however. Hill said that it is still unclear whether the company will have access to enough data to generate statistically valid models. While GNS has a good handle on its data requirements for preclinical research, the clinical realm is new territory, he noted. Ideally, "the more data, the smaller error bars you have," he said, but as for whether 20 to 30 patients per year will be enough, "we'll find out." So far, he said, GNS has performed some "theoretical tests" on synthetic data, which indicate that it is on the right track, but that has yet to be proven on real patient data.

Another unknown, Hill said, is the effect of patient variability on network modeling. "Clearly this is going to evolve," he said. "We can't take what worked in preclinical and apply it directly here — it's not a straight shot."

Nevertheless, he noted, "this is the first time that these kinds of advanced computational methods have been applied to clinical data." While acknowledging that some have "questioned how well it will work," he said, "I'm confident that it will be better than what they're using now." The company's ultimate goal in the project, he said, is to determine whether biosimulation "can impact patient care in real time."

Hill noted that despite the increased use of molecular profiling technology in cancer research, cancer treatment is still largely a "trial-and-error" process that has left "a huge gap between what can be done and what's done in practice." He said that recent trends may alter that situation, however, as physicians are under pressure from insurers to find patient populations that will respond to specific drugs.

"Insurers will stop [paying] if they're only getting 10-15 percent efficacy for a treatment that costs hundreds of thousands of dollars," he said.

Hill said that GNS is "in discussions" with other cancer centers about the possibility of using its modeling technology to classify patients in additional clinical trials, but again stressed that the collaboration with the Crowley Center is "still experimental."

Going forward, he said, "there's going to be a lot of hard slogging, but I think this approach will ultimately prove valuable."

— Bernadette Toner ([email protected])

Filed under

The Scan

Study Finds Few FDA Post-Market Regulatory Actions Backed by Research, Public Assessments

A Yale University-led team examines in The BMJ safety signals from the US FDA Adverse Event Reporting System and whether they led to regulatory action.

Duke University Team Develops Programmable RNA Tool for Cell Editing

Researchers have developed an RNA-based editing tool that can target specific cells, as they describe in Nature.

Novel Gene Editing Approach for Treating Cystic Fibrosis

Researchers in Science Advances report on their development of a non-nuclease-based gene editing approach they hope to apply to treat cystic fibrosis.

Study Tracks Responses in Patients Pursuing Polygenic Risk Score Profiling

Using interviews, researchers in the European Journal of Human Genetics qualitatively assess individuals' motivations for, and experiences with, direct-to-consumer polygenic risk score testing.