Central pattern generators drive rhythmic behaviors such as walking, chewing and flying. While they have been intensively studied, the underlying mechanisms are as yet not well understood. The long term objective of this search is to elucidate the neural mechanisms underlying the behavior of central pattern generators, in particular, the lobster stomatogastric ganglion (STG). Our approach is twofold: We will complement experimental investigation with modeling. Currently, there are a myriad of cell physiology parameters that could be incorporated into a model of circuit behavior (Selverston, 1988). By building simplifying models of the gastric mill portion of the STG, our goal is to determine which physiological parameters am necessary to account for the behavior, and which are peripheral to that explanation. My approach is to start with a minimal set, try to account for normal oscillatory behavior, and add properties as they appear necessary for accounting for more behavior. The modeling work also raises experimental questions which need to be answered to inform, constrain and validate the model. The modeling goals are: 1) Extend our current model with further constraints: We now can incorporate gap junctions, delay, and intrinsic currents in the model. We will use actual data to train the model instead of artificially produced sine wave data to further constrain it. 2) Explore the ability of the model to predict cell properties where not every cell is known. This is known in other fields as a systems identification task. If successful this could be extremely useful technique for identifying circuit components that are inaccessible experimentally. 3) Investigate hypotheses regarding constraints concerning functional constraints on the form of circuit. 4) Investigate the relationship between weights and phase relationships. 5) Investigate analytical methods such as bifurcation theory and models of relaxation oscillators to understand the mathematical aspects of our model. These modeling experiments require more data than is currently available. following biological experiments as planned: 1) Obtain better synaptic strength and electrotonic coupling estimates under normal and neuromodulated conditions. 2) Obtain a better estimate of the input/output function and intrinsic currents of a single neuron under different conditions. 3) Determine the effect of perturbations caused by cell kills and hyperpolarization on the output of the gastric mill. 4) Determine the changes m non-spiking oscillations following exposure to putative neuromodulators.
| Tsung, F S; Cottrell, G W (1995) Learning in recurrent finite difference networks. Int J Neural Syst 6:249-56 |
