Fig. 8: Transition rates estimation for data generated with the patch-clamp setup.

A The experimentally derived step response of the patch-clamp setup considerably deviates from the simulated step response of a 4-pole low-pass Bessel filter. B Comparison of the power spectra of simulated and recorded noise. Noise has been recorded with different resistors of a cell model (blue: bath resistor, orange: patch resistor) and a real cell in the cell-attached configuration (brown), with the red arrows indicating stray noise. The simulated noise was generated with either white noise filtered with a digital 4-pole low-pass Bessel filter (cyan) or with the power spectrum of the corresponding recorded noise using the ³¹ algorithm (see methods, green: bath resistor, red: patch resistor, lime: cell-attached). C, D In total, eight NNs have been trained, each with a dataset containing a combination of a step response and noise type as indicated below the graph (Table 1 datasets No. 7–10 and 11–14, COCOC topology, with SNR = 4 to SNR = 6 and SNR = 8 to SNR = 10, for (C, D), respectively). Two sets of 100 time series with an SNR ≈ 6 and SNR ≈ 8 were recorded on the patch-clamp setup using ideal time series as voltage command protocols and the 2D-histograms were analysed with the respectively trained NNs. The diamond marks the median as well as the 25 and 75 percentiles, while the whiskers denote the 10 and 90 percentiles. Significance was tested with the Kruskal–Wallis-Test and pairwise with Dunn’s Test (**p < 0.01 and ***p < 0.001). The orange stripes indicate the predictions of the NNs on the test data that were not recorded on the patch-clamp amplifier, but was instead generated in the same way as the training data (Table 1 dataset No. 7 and 11 for (C, D), respectively). Note, the performance of the other NNs on their respective test datasets (Table 1 datasets No. 8, 9, 10 and No. 12, 13, 14 for (A) and (B), respectively) were very similar (red lines). The green stripes indicate the error scores for randomly drawn rates. The lines denote the medians and the boundaries of the stripes the 25 and 75 percentiles.