Back to Building a cell the easy way
Back to Creating large networks with GENESIS
Up: Table of Contents

Making more realistic cell models

Outline

Why bother?

Read Jim Bower's thoughts on choosing the level of detail to use in modeling in BoG Chapter 11.

How important is spike latency and initial interspike interval (ISI) vs. the final ISI in determining the behavior of a network of neurons that display spike frequency adaptation?

Current clamp experiments often show results similar to the recording below (from Izhikevich E.M. (2006) Dynamical Systems in Neuroscience. The MIT Press). Is it necessary that a model neuron used in a large network accurately fit the timing of these action potentials? Or will the large variation in properties of individual neurons somehow "wash out" these details in the network? This is still an open question that may be answered by further modeling studies. However, there are indications that this variable spike timing can significantly affect network behavior.

Here T₀ is the spike latency, or time between the application of the injection pulse and the first spike. Under conditions of low excitation, this could act as an additional propagation delay, and affect the behavior of the network. The increasing interspike intervals T₁ - T₅ can also affect the behavior of the network. Under conditions of high excitation, when the neuron is firing continously, the later ones will be more relevant than the early ones.

Spike frequency adaption may also be used as a mechanism for processing behaviorally relevant stimuli in the presence of many other sources of synaptic input. For example , Benda et al. have presented evidence that spike frequency adaption is used as a high pass filter to separate transient signals from slower oscillatory signals in the electrosensory system of weakly electric fish. (Benda, J., Longtin, A. and Maler, L. (2005) Spike-frequency adaptation separates transient communication signals from background oscillations, J. Neurosci. 25: 2312-2321)

Building the Model

The process of building a biologically realistic compartmental model of a neuron involves three steps:

1. Build a suitably realistic passive cell model, without the variable conductances. This subject is treated briefly in the Introduction to GENESIS section on Constructing the passive cell, and in detail in the Advanced Tutorial on Realistic Single Cell Modeling

2. Add voltage and/or calcium activated conductances. See BoG Chapter 7 for an overview of the various types of ionic conductances, such as calcium conductances, calcium-activated potassium conductances, and inactivating potassium conductances, and how they affect firing properties

3. Tune the model to better fit passive properties and channel parameters that are known only approimately from experiment. Chapter 7 of the BoG describes how the cell and channel editing features of Neurokit may be used to perform manual parameter searches. The documentation for The GENESIS Parameter Search Library and the example scripts in genesis/Scripts/param describe powerful methods for performing automated parameter searches in GENESIS. The Advanced Tutorial on Parameter Searching tools in GENESIS gives a good overview of parameter searching and a discussion of the issues involved, suggestions, hints, and pitfalls.

The Advanced Tutorial on Realistic Single Cell Modeling examines the complete process of constructing a structurally realistic neuron model, using specific examples of modeling cerebellar neurons.