Since its launch in 2001, Ithaca, NY-based startup Gene Network Sciences has applied its proprietary simulation platform to modeling sub-cellular systems, with its crowning achievement so far being a cancer cell model that includes more than 500 genes and proteins in over 2,000 states and 1,400 parameters.
But the company says its technologies, which include its Diagrammatic Cell Language and VisualCell simulation software, are capable of much more. As testament to that potential, GNS recently embarked upon a collaborative project with researchers from Cornell University and the University of California, San Diego, to build an anatomically realistic three-dimensional canine heart model — a flying leap beyond the cell wall, and directly into the macroscopic world of organ-scale simulation.
GNS CEO Colin Hill said that the project presents the company with a key learning opportunity as it works to extend the scale of its current approach. “Modeling at the level of electrophysiology and the heart has been around for a very long time relative to some of the more recent molecular- and pathway-level modeling,” Hill told BioInform, “and as we try to extend our platform to more macroscopic levels, going from single cells to populations of cells, to populations of cells within a particular 3D geometry, there’s a lot that we … can borrow from the more mature effort in modeling cardiovascular systems.”
Hill stressed that the four-year project, supported by a $2 million NIH grant, is basic research, with the goal of “getting a deeper understanding of ventricular fibrillation first and foremost.” However, with some estimates of the market for cardiovascular drugs approaching $77 billion by 2005, Hill said the company recognizes the “tremendous market opportunity” for the heart model, despite the “many different technology steps that we’ll need to get through before we start accessing those markets.”
Hill said that a particularly promising area for a predictive heart model is in cardiac toxicity, where drug candidates would be tested in silico to determine whether they induce arrhythmia. “We believe that a model that will eventually incorporate signaling pathways and other things will be very useful at better understanding — and better predicting, ultimately — which drugs will be safe for the heart,” Hill said.
But Hill noted that the days of commercializing a predictive heart model are still several years down the road. In the meantime, GNS and its collaborators must overcome a number of challenges in creating an organ-level 3D model out of molecular-level experimental data.
Cornell’s Robert Gilmour, principal investigator of the project, will provide the data — patch-clamp measurements of ionic currents from three types of cells found in the canine ventricle. At GNS, project leader Jeff Fox, who previously worked in Gilmour’s lab, will use the company’s technology to create ionic models based on the Cornell data. These models will be passed on to Wouter-Jan Rappel at UCSD, who will integrate the GNS ionic models with models of intracellular calcium dynamics. Rappel has already begun work on a 3D canine ventricle model — albeit with much less electrophysiology data than in the current project — and will assemble the components into the final model.
Gilmour noted that this project is not the first attempt to create ionic models for the heart, but that previous approaches ran aground when faced with the complexity of scaling from a single cell through to the organ level. “You very quickly get into some programming issues in terms of how are you going to handle all of these different kinds of cells, and the different orientations that they have, and the complexity of the anatomy,” he said. “You can do it with brute force, but even that isn’t very useful, even if you have a huge computer.”
Gilmour said that he expects GNS to provide better algorithms than have previously been used for handling the complexity of the data, while Rappel’s expertise in anatomic-scale modeling will address some of the hurdles in converting the two-dimensional “sheets” of cardiac cells that Gilmour has already assembled into three dimensions.
This is an extremely compute-intensive step. Fox said that simulating a single 2D sheet of cardiac tissue took up to a week on an eight-processor cluster, and the computational requirements of modeling in the third dimension will be enormous. To meet the demand, GNS is planning on increasing its computational capacity over the next few months, Fox said, and Rappel will have access to the Pittsburgh Supercomputer Center as well.
If the project progresses as planned, Gilmour said the model will be able to accurately predict the effects of drug candidates that target specific ion channels, and in doing so will surpass current cardiac simulation efforts, which are limited to recreating the behavior of the heart.
Hill said that the biggest challenge — and the biggest opportunity — for the project would be in integrating the different scales required to predict these macroscopic effects of microscopic perturbations. “Ultimately, the drug at the end of the day interacts at the molecular level,” he said, “and even though the effects are eventually at the tissue or organ level, one does need to be able to really cross and span these length scales and these time scales, and do that in a rigorous manner” — an objective, he said, “which is no small task.”