Application of Machine-Learning Methods to Recognize mitoBK Channels from Different Cell Types Based on the Experimental Patch-Clamp Results

Abstract

(1) Background: In this work, we focus on the activity of large-conductance voltage- and Ca2+-activated potassium channels (BK) from the inner mitochondrial membrane (mitoBK). The characteristic electrophysiological features of the mitoBK channels are relatively high single-channel conductance (ca. 300 pS) and types of activating and deactivating stimuli. Nevertheless, depending on the isoformal composition of mitoBK channels in a given membrane patch and the type of auxiliary regulatory subunits (which can be co-assembled to the mitoBK channel protein) the characteristics of conformational dynamics of the channel protein can be altered. Consequently, the individual features of experimental series describing single-channel activity obtained by patch-clamp method can also vary. (2) Methods: Artificial intelligence approaches (deep learning) were used to classify the patch-clamp outputs of mitoBK activity from different cell types. (3) Results: Application of the K-nearest neighbors algorithm (KNN) and the autoencoder neural network allowed to perform the classification of the electrophysiological signals with a very good accuracy, which indicates that the conformational dynamics of the analyzed mitoBK channels from different cell types significantly differs. (4) Conclusion: We displayed the utility of machine-learning methodology in the research of ion channel gating, even in cases when the behavior of very similar microbiosystems is analyzed. A short excerpt from the patch-clamp recording can serve as a "fingerprint" used to recognize the mitoBK gating dynamics in the patches of membrane from different cell types.

Keywords: K-nearest neighbors algorithm; autoencoder; gating dynamics; machine learning; mitoBK channels.

Figure 1

(Upper): the samples of the original ionic currents of single mitoBK channels for three different cell types: endothelium (blue); fibroblast (gray); hippocampus (red). (Lower): the average probability of channel opening at different membrane potentials fitted by sigmoidal curves.

Figure 2

An exemplary visualization of the recordings reduced to a 2-dimensional space with the autoencoder corresponding to different pipette potentials and different cell types. Each dot represents a 1000-data-points-long subseries from the original recording. Some of the dots overlap each other.

Figure 3

The set of confusion matrices for different pipette potentials. The values given in each block of the confusion matrices indicate the number and percentage of samples classified in a given category. Please note that the overall number of samples for each potential $U_m$ is different, but it is equally distributed among three cell lines.

Figure 4

The scheme presenting the subsequent stages of the patch-clamp data processing.

Figure 5

(a) Illustration of the time series split. In the current analysis the $r=200$. (b) A schematic representation of the PAA method. Each original subseries of 1000 data points (which corresponds to the time interval of 0.1 s at the applied sampling rate of 10 kHz) is “compressed” to the series of 200 points by the PAA algorithm of dimensionality reduction. Here, for clarity we presented the dimensionality reduction performed on the original subseries of 200 points (corresponding to the time interval 0.02 s). As one can see, application of PAA algorithm allow the retention of the main features of the original input series.

Figure 6

The scheme of autoencoder. The green squares stand for the input data, the orange ones indicate the intermediate layer of neurons. The red squares denote the latent space and provide the reduced representation of the sample.

Figure 7

The representation of the applied technique of anomaly detection. The plot at the bottom shows the representation of the data corresponding to the fibroblast cell line at 40 V in 2-dimensional space after its compression with the autoencoder. Different colors in the figure correspond to the independent patches of the same cell line. Each dot represents a 1000-data-points-long subseries from the original recording. As one can see, within the recordings represented by the dark blue dots and the orange ones there were some intervals where the activity of the channel stands out from its typical characteristics represented by the cluster at the top (dark blue) and some orange dots at the left bottom. These outstanding points are called here “outliers”. The plots on the top illustrate the reconstruction error dependence on the dimension of the latent space for positive (typical) and outlier data, respectively.

Figure 8

(a) Representation of K-nearest neighbors algorithm: general scheme of algorithm, (b) illustration of confusion matrix.