A chi(2) test for model determination and sublevel detection in ion channel analysis.

A chi(2) test is proposed that provides a means of discriminating between different Markov models used for the description of a measured (patch clamp) time series. It is based on a test statistic constructed from the measured and the predicted number of transitions between the current levels. With a certain probability, this test statistic is below a threshold if the model with a reduced number of degrees of freedom is compatible with the data. A second criterion is provided by the dependence of the test statistic on the number of data points. For data generated by the alternative model it increases linearly. The applicability of this test for verifying and rejecting models is illustrated by means of time series generated by two distinct channels with different conductances and by time series generated by one channel with two conductance levels. For noisy data, a noise correction is proposed, which eliminates noise-induced false jumps that would interfere with the test. It is shown that the test can also be extended to aggregated Markov models.