Simulation and modeling of neurobiological networks with applications to cerebellar networks implicated in classical conditioning of the eyeblink response

Classical conditioning of the eyeblink response in rabbit consists of repeated pair-wise presentations of a tone, and an airpuff blown into the eye. The airpuff always elicits an eyeblink. Eventually, conditioning takes place, i.e., the rabbit learns to blink in response to the tone alone. Neurobiological studies have implicated the cerebellar cortex and the cerebellar nuclear region, the interpositus, as neuronal substrates where the memory trace for eyeblink conditioning resides. Synaptic plasticity in the same cerebellar areas is implicated as the mechanism for learning.;Two approaches to neural network modeling are pursued. The first involves a network of leaky integrator neurons, and is based on using a system identification algorithm to learn the weights that result in a close match between model outputs and target data. The comparison of the weights resulting from the network matching first before-conditioning data, and then after-conditioning data yields the model's predictions of plasticity sites in the network. The likelihood of the prediction is increased by means of plasticity site identification using the duplex network, a combination of the before- and after-conditioning networks that provides for a method of selective clamping of weights that avoids pre-assignment of values.;The second approach consists of compartmental modeling using the neural network simulator CAJAL. A parameter estimation method is used in conjunction with the SIBN simulation environment to likewise match before- and after-conditioning data and to produce plasticity hypotheses. Our models strongly suggest parallel fiber-Purkinje and pontine nucleus-interpositus synapses as the sites as plasticity.;The central proposition of this thesis is that through simulations of neural network models, it is possible to determine with a high degree of likelihood the sites of synaptic plasticity in the implicated neural circuitry, and further to predict the direction of plasticity (i.e., whether the synapse is strengthened or weakened) in each case. Our models are based on the neural circuitry in the cerebellar cortex and nuclei, and are free of built-in predisposition regarding the sites of plasticity.