Applicant
PD Dr. Tobias Huth
Institut für Physiologie und Pathophysiologie
Friedrich-Alexander-Universität Erlangen-Nürnberg
Project Overview
This project aims to optimize and deploy a simulation-based deep-learning framework for the analysis of single-channel patch-clamp recordings of ion channels. Ion channels are crucial for electrical signaling in almost all organ systems, contributing to normal physiological function, while alterations in their behavior underlie a wide range of pathophysiological conditions. Ion channels stochastically switch between closed and open states, a process termed gating. The single-channel patch-clamp method records the electrical current through a single ion channel protein complex [1], and quantitative kinetic models derived from such recordings provide a direct link between molecular channel structure and function.
We devised an algorithm that relies on large-scale stochastic simulations of ion-channel gating, computed as a continuous-time Markov process using Gillespie’s stochastic simulation algorithm [2], to generate training data for neural networks [3], building on the parallel simulation pipeline established for high-performance computing systems [4]. This approach enables inference of ion-channel models directly from experimental recordings in near real time. The core challenge addressed here is the extreme computational cost arising from the large number of simulations required and the computational demands of realistic noise modelling, which currently limits scalability and broader applicability. This substantial computational effort is rewarded by enabling the investigation of biophysical mechanisms, drug-channel interactions, and disease-related phenotypes driven by altered channel kinetics.
Within the framework of KONWIHR, we will (i) redesign the simulation pipeline for hybrid CPU-GPU execution, including GPU offloading of the inverse Fast Fourier Transform (IFFT) computations used for noise generation, (ii) implement on-the-fly simulation-training workflows to minimize I/O and storage overhead, and (iii) optimize resource efficiency and usability for deployment on both HPC systems and smaller workstation infrastructures. The figure below illustrates this on-the-fly workflow and the allocation of the individual simulation subtasks to CPU and GPU. The project will result in software that enables more efficient and automated analysis of ion-channel data for the wider scientific community.
References
[1] Sakmann, B., & Neher, E. (Eds.). (1995). Single-Channel Recording (2nd ed.). Plenum Press.
[2] Gillespie, D. T. (1977). Exact stochastic simulation of coupled chemical reactions. The Journal of Physical Chemistry, 81(25), 2340–2361. https://doi.org/10.1021/j100540a008
[3] Oikonomou, E., Juli, Y., Kolan, R. R., Kern, L., Gruber, T., Alzheimer, C., Krauss, P., Maier, A., & Huth, T. (2024). A deep learning approach to real-time Markov modeling of ion channel gating. Communications Chemistry, 7(1), 280. https://doi.org/10.1038/s42004-024-01369-y
[4] Oikonomou, E., Gruber, T., Chandra, A. R., Höller, S., Alzheimer, C., Wellein, G., & Huth, T. (2023). 2D-dwell-time analysis with simulations of ion-channel gating using high-performance computing. Biophysical Journal, 122(7), 1287–1300. https://doi.org/10.1016/j.bpj.2023.02.023