Creating mechanistic ion channel models using genetic algorithms

Modeling ion channel function is often problematic, especially for channels that have not been physically characterized, because transitions between many conformational states cannot be directly observed. Myriad voltage clamp protocols have been developed to study these transitions, but model parameterization remains difficult because either all transitions must be culled from one protocol, or the data from multiple protocols must be non-redundantly combined. One solution to this problem is to use a genetic algorithm to create models that fit all desired voltage clamp data. This method has the distinct advantage of producing accurate ion channel models without manipulating the voltage clamp data in any way that might introduce artifacts, biases, or errors. I have developed software that not only parameterizes but creates the structure of Markov (state-based) models using this method. To determine whether this method can structure and parameterize mechanistically accurate models, a test model was created that approximately represented a voltage-gated sodium channel and was used to obtain simulated voltage clamp data using two protocols. The genetic algorithm was then used to create and parameterize models whose simulated output closely approximated the test data. The resulting models not only accurately approximated the test data, but their structure and parameter values were similar to those of the test model. This method is therefore well suited not only for parameterization of Markov models, but for model selection.