October 14, 2004 | If a rose by any other name is still a rose, what is Molecular Epistemics? Explaining the concept and corresponding proprietary technology platform is the challenge facing Genstruct, a young systems biology company founded in 2001. If you divide the systems biology world into two broad camps — wet-lab-focused data generators and in silico model makers — Genstruct is a modeler with a twist.

"Instead of using mathematical relationships to model the interactions between different entities, we use logic relationships," says Keith Elliston, Genstruct co-founder, CEO, and president. "We think of the world in terms of state changes. We can take any kind of state change — gene-expression change, protein change, metabolite change, to phenotypic change — and we can infer sets of causes and consequences related to those changes."

Elliston contrasts Genstruct's approach with the discrete mathematical modeling approach of other systems biology firms such as Entelos, which might model a specific phenotypic effect (e.g., serum glucose levels) after treating with a drug at x concentration. "Our approach would be to say, 'Give me the 500 genes that changed when you treat this patient with glucose, and I'll tell you the series of interactions that had to take place from a signaling perspective to explain exactly the mechanism by which glucose will increase.'"

Brash aspirations are endemic in the world of systems biology. So far, big results are scarce. Mindful of industry misgivings, Genstruct is building a seasoned team and hoping its brand of systems biology will attract prominent paying collaborators.

Keith Elliston,  Genstruct Co-Founder, CEO and President

Elliston was formerly Merck's director of bioinformatics and scientific director of the Merck Gene Index project. He also had stints at Bayer, Gene Logic, and he founded Viaken Systems. He founded Genstruct with Noubar Afeyan, CEO of Flagship Ventures, and the late Navin Chandra, former Perot Systems vice president of technology and expert in knowledge representation.

Doug Lauffenburger, a systems biology and bioengineering authority from MIT, is on Genstruct's board of directors, and Joshua LaBaer, director of Harvard's Institute of Proteomics, is on the scientific advisory board. Now Genstruct must put its impressive brain trust to work.

Elliston has sufficient cash to operate comfortably into late 2005 ($6.5 million in Series A financing in September 2003). There's plenty of room to house the staff of 21 in the new Cambridge, Mass., digs, and to grow by 10 in the next 12 months. In mid September, Genstruct entered a deal with Berlex, Inc., a U.S. affiliate of Schering AG, to characterize the regulatory mechanisms for breast cancer. The goal is to uncover the causal mechanisms for the initiation of breast cancer to better understand the genes, proteins and metabolites that play crucial roles in the development of this disease.


Knowledge Assembly
The real question is: Will the company's technology be able to uncover new mechanisms of action (MOAs) as promised? Here's the approach: So-called "biological case frame" templates — Genstruct's proprietary representation of biology — are combined to make "knowledge assembly" models. These models are customized for a specific system (e.g., diabetes) and interrogated on the Epistemics engine to identify novel MOAs, which may suggest targets and therapeutics.

So far, Elliston says, the company has three well-developed models:

• Type II diabetes — regulation of glucose transfer, glucose metabolism, insulin signaling, fatty acid synthesis
• Oncology — cell cycle regulation, cancer mechanisms
• Dyslipidemia — lipid metabolism, cholesterol metabolism, transport, and their regulation

A fourth effort around inflammation is just kicking off. The company has filed for eight patents. The recent deal with Berlex should help build credibility.

"Now we will work with Genstruct to define the cascade of molecular actions that lead to the onset and advancement of breast cancer and validate Genstruct's model of breast cancer through molecular tests in our laboratories. As a company that specializes in both oncology and women's healthcare, we at Berlex hope this collaboration will lead to new therapies that impact the disease earlier in its course," says Harald Dinter, Ph.D., scientific director of oncology research, Berlex.

In the meantime, Genstruct is trumpeting its analysis of two gene-profiling studies on type II diabetes populations*, one Scandinavian and one Mexican-American. It studied changes in 750 genes in the two studies. While the original researchers were able to rationalize only 15 of the changes, Genstruct was able to "explain" 250 changes primarily on genes related to oxidative phosphorylation in mitochondria.

Jack Pollard, Genstruct's project leader for metabolic disorders, presented this work at the American Diabetes Association's annual meeting last spring.

"A lot of gene-expression analyses are largely cataloging exercises — no offense intended, but they typically do some differential profiling of the diseased state versus normal state and use a myriad of approaches, be it clustering or representation analysis, to reconstruct gene regulatory networks at some low level," Pollard says. "But at the end of the day, they haven't explained consequence from cause in terms of cause of the disease and the cause of the expression profile.

"We used our model of human skeletal muscle biology in type II diabetes to analyze the data set. The conclusion was a set of hypotheses about calcium signaling that explain a lot of gene-expression changes. We were able to explain why those changes in oxidative phosphorylation genes might be related to decreases in insulin receptor signaling."

Testing its hypotheses in the lab is a necessary (and iterative) step. Currently Genstruct doesn't do wet-lab work, relying instead on partners. The IT required to churn out sophisticated in silico models is surprisingly vanilla: lots of "commodity" Unix machines (60 to 100 CPUs) and about 10 TB of disk storage, Elliston says. Much of the programming is in Java and C++, and the Molecular Epistemics platform was built on Web Services. A big relational database underpins the platform.

What's needed now is a big project — and result — to prove to potential collaborators that Genstruct's hypotheses creation engine is more than just an intriguing idea.