Deciphering Ion Channel Dynamics: A Clustering Approach for Signal Analysis

This thesis investigates clustering techniques for analyzing multi-channel CFTR ion channel activity recorded via the patch-clamp method. Accurately determining the number of active channels and classifying their states is crucial for understanding ion channel dynamics. Both classical and machine learning approaches are explored. Classical methods, including DBSCAN and a hybrid DBSCAN-BGMM approach, show limitations in handling noise and overlapping states, particularly with increasing channel counts. A novel machine learning approach combining a Cluster Count Neural Network (NN) for channel number estimation and a Long Short-Term Memory (LSTM) network for state classification demonstrates superior performance. The NN effectively captures underlying patterns while the LSTM leverages temporal dependencies, achieving higher accuracy and robustness even with complex, noisy signals. This research offers promising tools for analyzing ion channel activity and has implications for cystic fibrosis research and drug development.