| Computational methods were used to investigate the generation and modulation of feeding behavior in the mollusc Aplysia californica. In the first set of experiments, feed-forward neural networks were trained with back-propagation. Analogues of neuronal stimulation and ablation were then used to simulate processes determining the functional roles of individual units. We found that neurons which at first appeared to be 'command- like', functioned in a more complex, distributed manner. This suggests that, in biological systems, it may not be possible, without the knowledge of the whole system, to infer the functional role of a single neuron from simple recording, stimulation and ablation methods. The second set of experiments used continuous time neural network simulations with genetic algorithms, to optimize parameters of a simple circuit that could produce rhythmic biting behavior. We then studied the circuits to determine whether modulatory mechanisms improve behavioral efficiency. We first observed that even in the absence of any plasticity or neuromodulation, the system could exhibit a “warm-up” or arousal effect. Change in one of the intrinsic neuronal properties (e.g. spike after potentials) affected the duration of warm-up, without altering certain parameters such as burst period. As predicted, in the absence of neuromodulatory mechanisms, these simple networks did not function well when challenged by a variety of internal or external alterations, in particular, conditions that resulted in an increase in the period of the program. However, it was found that with the addition of a cotransmitter mechanism that simulated peptidergic intrinsic modulation seen in real animals, the system autoregulated over a range of program periods. These results support the idea that cotransmitters may have evolved to maintain effective functioning of a neuromuscular system that must operate under varying conditions. |
