Atomic scale analysis of conduction events in potassium channels through neural networks

Ion channels are essential for various physiological processes, as they regulate the movement of ions across cell membranes. In recent years, Molecular Dynamics (MD) and Markov State Models (MSMs) have been used to study these processes at the atomic scales. One important limit of the methods currently used to estimate MSMs from MD trajectories of conduction events is related to the clustering step, which is complicated by the high dimensionality of the dataset. This thesis investigates the application of neural network models, specifically Variational Autoencoders (VAEs) and networks based on the Variational Approach for Markov Processes (VAMPnets), for the atomic-scale analysis of conduction events in potassium ion channels. Results indicate that both VAEs and VAMPnets can identify the macrostates conduction events in simulated data, but VAMPnets provide a superior ability to identify and characterize slower processes. Both models show limitations in replicating exact transition matrices particularly for systems with fast dynamics. This work demonstrates the potential of deep learning techniques to study ion channel dynamics.