RESUMO
The algorithm proposed here for automatic level detection in noisy time series of patch-clamp current is based on the detection of jump-free sections in the time series. The detector moves along the time series and uses a chi(2) test for the detection of jumps. When a jump is detected, the mean value, the variance and the length of the preceding jump-free section are stored. A Student's t-test was employed for the assignment of detected jump-free sections to discrete levels of the Markov model and for rejection of all sections with multiple assignments. The choice of the two significance levels is based on a 3-D diagram displaying the average number of detected levels from several time series vs. the significance levels of jump detection and of level assignment. The correct one is selected out of several plateaus with integer number of levels by means of the criterion of minimum scatter or other plausibility considerations. The test has been applied to simulated data obtained from a 2-state model and a 5-state aggregated Markov model, and the influences of SNR and of gating frequency are shown. Finally, the performance of the level detector is compared with a fit-by-eye and with a fit of the amplitude histogram by a sum of gaussians. At high noise, the fit of amplitude histograms failed, whereas the other two approaches were about equal.
Assuntos
Algoritmos , Simulação por Computador , Canais Iônicos/fisiologia , Modelos Biológicos , Técnicas de Patch-Clamp/métodos , Eletrofisiologia/métodos , Eucariotos/fisiologia , Ativação do Canal Iônico/fisiologia , Cadeias de Markov , Potenciais da Membrana/fisiologia , Modelos Estatísticos , Distribuição Normal , Controle de Qualidade , Sensibilidade e Especificidade , Processos EstocásticosRESUMO
Exact algorithms for the kinetic analysis of multichannel patch-clamp records require hours to days for a single record. Thus, it may be reasonable to use a fast but less accurate method for the analysis of all data sets and to use the results for a reanalysis of some selected records with more sophisticated approaches. For the first run, the tools of single-channel analysis were used for the evaluation of the single-channel rate constants from multichannel dwell-time histograms. This could be achieved by presenting an ensemble of single channels by a "macrochannel" comprising all possible states of the ensemble of channels. Equations for the calculations of the elements of the macrochannel transition matrix and for the steady-state concentrations for individual states are given. Simulations of multichannel records with 1 to 8 channels with two closed and one open states and with 2 channels with two open and two closed states were done in order to investigate under which conditions the one-dimensional dwell-time analysis itself already provides reliable results. Distributions of the evaluated single-channel rate constants show that a bias of the estimations of the single-channel rate constants of 10 to 20% has to be accepted. The comparison of simulations with signal-to-noise ratios of SNR = 1 or SNR = 25 demonstrates that the major problem is not the convergence of the fitting routine, but failures of the level detector algorithm which creates the dwell-times distributions from noisy time series. The macrochannel presentation allows the incorporation of constraints like channel interaction. The evaluation of simulated 4-channel records in which the rate-constant of opening increased by 20% per already open channel could reveal the interaction factor.