GLIF Models

Generation of detailed biophysical data through standardized, systematic experimental methods facilitates the creation of computational models that simulate or predict cell behavior. The data created as part of the Allen Cell Types Database can be used in multiple different types of models and simulations. For simulations of neural networks, there is a tradeoff between the size of the network that can be simulated and the complexity of the model used for the individual neurons. A series of models of increasing complexity was constructed to reproduce the spiking behaviors of the recorded mouse and human neurons. Starting with a leaky integrate-and-fire model, three generalizations were added: a) after-spike currents which represent the slower effects of ion channels activated by an action potential, b) subthreshold voltage and spike-dependent changes in threshold caused by the activation and inactivation of ion channels and c) voltage and threshold reset rules derived, from the electrophysiology data. These rules determine how the threshold and voltage are reset after a spike and depend on the state prior to the action potential. Electrophysiological stimuli were specifically designed to estimate some of the parameters of the generalizations. Following these initial estimates, a threshold parameter was further tuned to optimize the reproduced spike times generated by a training noise stimulus. The optimization method was based on maximizing the likelihood of a model neuron with intrinsic noise exactly reproducing the spike train observed in the experiment. The model performance was subsequently evaluated on a test stimulus: for different time scales, the fraction of the variance of the neuronal response which was explained by the model was computed.