Project Details
Projekt Print View

Single-channel Markov modelling of voltage-gated ion channels with simulations and implementation of the 2D-Fit algorithm on High Performance Computing Cluster

Subject Area Cognitive, Systems and Behavioural Neurobiology
Biophysics
Molecular Biology and Physiology of Neurons and Glial Cells
Term from 2020 to 2023
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 422920516
 
Final Report Year 2024

Final Report Abstract

Single-channel patch-clamp recordings provide a wealth of information about the function of ion channels and, in principle, allow for more accurate modeling of ion channel kinetics compared to macroscopic whole-cell recordings. Our goal was the extraction of a hidden Markov model (HMM) from a recorded time series, which are typically used to describe ion channel kinetics. Analyzing single-channel patch-clamp recordings is endowed with inherent difficulties stemming from a low signal-to-noise ratio (SNR), since the recorded signal originates from a single channel protein with electrical currents in the range of fA to pA. This makes low-pass filtering the signal indispensable leading to a reduction of the effective recording bandwidth that impede analytical modeling of the data. We employed the 2D-Fit algorithm that approximates the experimentally recorded data with simulations and iteratively derives the model parameters. The 2D-Fit has now been upgraded to be compatible with High Performance Computing (HPC) clusters for massive parallel computing, allowing ensemble solutions to be obtained. Using these resources, we demonstrated the importance of ensembles for challenging tasks and the superior performance of the 2D-Fit related to noisy backgrounds, fast gating events beyond the corner frequency of the low-pass filter, and topology estimation of the underlying HMM. We are now in the process of employing the 2D-Fit to real experimental time series of voltage gated ion channels. The first milestone has already been achieved. We have already successfully applied the algorithm to patch-clamp time series of the MaxiK potassium channel. We obtained robust HMMs with the 2D-Fit that generate time series almost indistinguishable from recorded time series. Additionally, we developed a Deep Learning (DL) pipeline for ion channel modeling. The 2D-DL approach is based on the same principles as the 2D-Fit. However, the simulations are conducted prior to modeling the experimental time series in order to generate synthetic training data. After the neural networks are trained, this approach has fundamental advantages. The computational burden to simulate the training data and to train the networks is a onetime endeavor. Therefore, the resource cost is drastically reduced compared to the 2D-Fit. There is already a strong indication that with trainable filters in the input stage of the neural networks, the modeling performance can be further improved. In the future, this approach would allow modeling in quasi-real-time during an ongoing patch-clamp recording session.

Publications

 
 

Additional Information

Textvergrößerung und Kontrastanpassung