Modelling excitable cells using cycle-linear hybrid automata.

Cycle-linear hybrid automata (CLHAs), a new model of excitable cells that efficiently and accurately captures action-potential morphology and other typical excitable-cell characteristics such as refractoriness and restitution, is introduced. Hybrid automata combine discrete transition graphs with continuous dynamics and emerge in a natural way during the (piecewise) approximation process of any nonlinear system. CLHAs are a new form of hybrid automata that exhibit linear behaviour on a per-cycle basis but whose overall behaviour is appropriately nonlinear. To motivate the need for this modelling formalism, first it is shown how to recast two recently proposed models of excitable cells as hybrid automata: the piecewise-linear model of Biktashev and the nonlinear model of Fenton-Karma. Both of these models were designed to efficiently approximate excitable-cell behaviour. We then show that the CLHA closely mimics the behaviour of several classical highly nonlinear models of excitable cells, thereby retaining the simplicity of Biktashev's model without sacrificing the expressiveness of Fenton-Karma. CLHAs are not restricted to excitable cells; they can be used to capture the behaviour of a wide class of dynamic systems that exhibit some level of periodicity plus adaptation.

Efficient event-driven simulation of excitable hybrid automata.

We present an efficient, event-driven simulation framework for large-scale networks of excitable hybrid automata (EHA), a particular kind of hybrid automata that we use to model excitable cells. A key aspect of EHA is that they possess protected modes of operation in which they are non-responsive to external inputs. In such modes, our approach takes advantage of the analytical solution of the modes' linear differential equations to eliminate all integration steps, and therefore to dramatically reduce the amount of computation required. We first present a simple simulation framework for EHA based on a time-step integration method that follows naturally from our EHA models. We then present our event-driven simulation framework, where each cell has an associated event specifying both the type of processing next required for the cell and a time at which the processing must occur. A priority queue, specifically designed to reduce queueing overhead, maintains the correct ordering among events. This approach allows us to avoid handling certain cells for extended periods of time. Through a mode-by-mode case analysis, we demonstrate that our event-driven simulation procedure is at least as accurate as the time-step one. As experimental validation of the efficacy of the event-driven approach, we demonstrate a five-fold improvement in the simulation time required to produce spiral waves in a 400-x-400 cell array.

A cycle-linear approach to modeling action potentials.

We introduce cycle-linear hybrid automata (CLHA) and show how they can be used to efficiently model dynamical systems that exhibit nonlinear, pseudo-periodic behavior. CLHA are based on the observation that such systems cycle through a fixed set of operating modes, although the dynamics and duration of each cycle may depend on certain computational aspects of past cycles. CLHA are constructed around these modes such that the per-cycle, per-mode dynamics are given by a time-invariant linear system of equations; the parameters of the system are dependent on a deformation coefficient computed at the beginning of each cycle as a function of memory units. Viewed over time, CLHA generate a very intuitive, linear approximation of the entire phase space of the original, nonlinear system. We show how CLHA can be used to efficiently model the action potential of various types of excitable cells and their adaptation to pacing frequency.

Hybrid automata as a unifying framework for modeling excitable cells.

We propose hybrid automata (HA) as a unifying framework for computational models of excitable cells. HA, which combine discrete transition graphs with continuous dynamics, can be naturally used to obtain a piecewise, possibly linear, approximation of a nonlinear excitable-cell model. We first show how HA can be used to efficiently capture the action-potential morphology--as well as reproduce typical excitable-cell characteristics such as refractoriness and restitution--of the dynamic Luo-Rudy model of a guinea-pig ventricular myocyte. We then recast two well-known computational models, Biktashev's and Fenton-Karma, as HA without any loss of expressiveness. Given that HA possess an intuitive graphical representation and are supported by a rich mathematical theory and numerous analysis tools, we argue that they are well positioned as a computational model for biological processes.