Login to Your Account

Ariana's Statistical Algorithms to Separate Signal from Noise


By Cormac Sheridan
Staff Writer

Ariana Pharma SA is developing software to guide the selection of personalized cancer care as part of a clinical trial in patients which, the study investigators claim, will be the first to offer all participants a treatment regimen based on their individual DNA, RNA and microRNA profiles.

The Winther Consortium, which is led by by MD Anderson Cancer Center in Houston, and Institut Gustave Roussy, of Paris, is recruiting 200 patients with advanced solid tumors at centers in the U.S., France, Canada, Spain and Israel. Data are expected in 2015.

All will be offered biologically guided therapies, whereas in typical cancer trials about 70 percent of patients are offered therapies based on standard protocols. The study protocol will include the use of dual biopsies, to enable comparisons between cancerous tissues and healthy tissues in each patient.

"All this talk about personalized medicine is gradually becoming a reality," Mohammad Afshar, CEO of Paris-based Ariana Pharma, told BioWorld Today. "Ultimately, the idea is to bring better treatment at that particular time to that particular patient with that particular tumor."

Ariana, which was spun out of the Paris-based Institut Pasteur 10 years ago, is bringing a statistical method called concept framework analysis – also known as Galois lattice analysis – to the interpretation of the biological data that will flow in the study.

That approach, Afshar said, enables easily interpreted analysis of multiple parameters.

"It's a tool to generate hypotheses, which can then be tested," he said. "The biologist can look at that and ask, 'Does this make sense? Can I link it to a particular pathway?'"

In personalized medicine – just as in drug development more generally – much of the low-hanging fruit already may have been picked.

The number of single genetic markers that have strong prognostic value, such as HER2/neu or KRAS, has remained limited. But building reliable panels of biomarkers with similarly strong predictive capabilities has proved, apart from a small number of exceptions, difficult. "The difference is when you start combining things," Afshar said. "You can end up with multiple answers which are statistically equivalent."

Part of the problem stems from the inadequacy of traditional biostatistical approaches, he said, which are based on analyzing a small number of descriptors or parameters in a large sample of patients. Modern molecular methods have led to an explosion in the number of data points that are generated per patient. "We have a fundamental mathematical problem called overfitting," Afshar said.

The result is often more noise than signal. Molecular signatures that appear to work in one particular setting often prove to be not generally applicable in wider patient populations. "That is really something which is plaguing the whole field," he said.

The so-called Winther trial will be the first trial to evaluate a clinical treatment algorithm developed by Ariana. However, the company's analytics capabilities already have been deployed in several diagnostic projects.

A collaborative project with Berkeley, Calif.-based Bio-Rad Laboratories Inc. will shortly lead to a new prognostic that will predict likely responders to therapy in hepatitis C infection. Further information will become available in the fourth quarter.

"Watch this space," Afshar said.

It is also part of a group that will shortly submit for publication a paper on predicting renal transplant rejection. And Ariana previously collaborated with Fovea Pharmaceuticals SA, now part of Paris-based Sanofi SA, on the analysis of Phase II data for the allergic conjunctivitis drug prednisporin.