Hidden Markov modeling of ion channel currents

The activity of ion channel proteins is central to many physiological processes. Modeling the behavior of ion channels from the observed single channel data gives important kinetic information and insights into their mechanism of activation and is useful in distinguishing among alternative molecular mechanisms in drug action or disease. The focus of this thesis is on the development and application of signal processing techniques for analysis of single channel current recordings at low signal to noise ratios when traditional algorithms such as threshold detection or histogram analysis cannot be applied.;The problem of kinetically modeling the protein has been approached in a hidden Markov model (HMM) framework. The problem is to find the maximum likelihood estimate of the parameters of the Markov model from the observed single channel data. The main contribution of this thesis is the adaptation of the traditional forward-backward and Baum-Welch algorithm to estimate the kinetic parameters of the HMM while realistically modeling the underlying signal and noise.;In practice, the additive background noise in a patch-clamp recording is correlated. Excess noise is also seen when the channel is partially-open or open. The problem of correlated noise is addressed by modeling the single channel data as the output of a vector or metastate hidden Markov process. A model for the state dependent and correlated noise is introduced and an algorithm for reestimating the noise model and HMM parameters is presented. Adaptation of the algorithms are also proposed to characterize filtered, sampled continuous time data and extended to address multiple observation sequences and baseline estimation.;The algorithms have been applied to data from a voltage-gated potassium channel. It is shown via simulations and from analysis on experimental data that the HMM theory promises to allow meaningful kinetic information to be obtained from channels whose characterization in the past have been limited by rapid kinetics or poor signal to noise ratios in the recordings.