This weekend at the Beyond Genome meeting in San Diego, Gene Network Sciences unveiled its computer simulation of a human colon cancer cell — the largest and most complex such model yet available, according to the company.
Packed with more than 2,000 variables for approximately 500 genes and proteins and loaded with a good portion of the signal transduction pathways and gene expression networks that control the mammalian cell cycle, the model is simply huge. By comparison, the most advanced cellular model developed to date has been Masaru Tomita’s E-Cell at Keio University, an in silico model of the Mycoplasma genitalium bacterium, which contains only 127 genes.
According to GNS CEO and founder Colin Hill, the GNS human colon cancer cell, which takes up 66 square feet of floor space when printed out in poster form, describes the processes of endocytosis, receptor signaling, protein degradation, signal transduction, transcriptional control of gene expression networks, and protein translation.
In addition, he said, it is able to predict physiological outcomes such as cell cycle progression, cell cycle arrest, and apoptosis induction, and should have applications in the study of other types of cancer as well as Alzheimer’s disease and arthritis.
The Ithaca, NY-based company has not yet published this work in a peer-reviewed journal, but plans to do so, Hill said.
Hill said GNS was able to build the highly complex model in just a year and a half using a number of proprietary computational methods, such as the diagrammatic cell language, developed specifically to represent cellular systems, and network inference methodology, which infers missing pathways and networks using known biological data. These efforts were substantially sped up in recent months, however, by access to a 192-processor Linux-based supercomputer that GNS is test-driving under an agreement with IBM.
While simulating cellular processes is computationally intensive, it pales in comparison to the number crunching necessary to add data to the model. Because so little of the cell’s circuitry is actually known — estimates run around 5 percent — GNS uses its network inference methodology in a highly parallel fashion to match the astronomical numbers of available parameters and network topologies with the given data. Even 192 processors, while a great help, “is not really enough,” Hill said. “We’ll outgrow that very quickly.”
GNS said it’s ready to pursue its strategy of using the predictive powers of the cellular model in collaborations with pharmaceutical and biotech companies. Of interest to these potential clients should be the model’s key contents — between a quarter and a half of all drug targets for cancer, by Hill’s estimate.
But how receptive will these potential customers be to such a nascent approach to drug discovery? Hill said that although GNS has yet to sign a deal, his discussions with prospective clients have been positive, largely because the company is very careful about validating all of its predicted results in its wet lab. Because of this focus, “it’s hard to discount or poke holes in our methodology,” he said. “We’re not coming to a pharma company with predictions that are just predictions.”
Of course, even state-of-the art technology has its limits. Although GNS said it expects to incorporate “every published and publicly known drug target for cancer” and “all of the [published] biology at the level of pathways” into its simulation within the next 12 months, that information still leaves 95 percent of the cell’s circuitry unaccounted for, a situation that Hill sees as the key opportunity for the future of systems biology. “There’s always going to be missing biology,” he said. Techniques such as GNS’ network inference methodology, if proven effective, will give the largely unproven field a significant boost.
“People are looking for this field to grow up and start having some impact in drug discovery and other areas,” said Hill. The future success of the company will certainly hinge on its ability to convince prospective customers that its model is a step in that direction.