Reconstruction of neuron potentials with convolutional neural networks trained on nanoelectrode recordings
Abstract
Neurons communicate through electrical impulses resulting from changes in their membrane potentials. The fine features of potential waveforms reflect the dynamics of several coordinated voltage-gated ion channels on the cell membrane, and any changes in these dynamics results in malfunctions of the brain. Therefore, electrophysiological recordings of neuronal potentials are essential in understanding connectivity and communication among neurons, revealing the biophysical principles underlying learning and memory, and to understand and treat neurological disorders. To obtain high resolution neuronal potentials that can resolve fine waveform features, current methods are invasive and require the measurement tool (electrode) to be inserted into the protective membrane of the cell. If the electrode is placed close to the cell but not inserted, a transformed waveform can be obtained which provides limited information about the ion channel dynamics, but in a minimally invasive manner. We have developed nano-meter scale electrodes (nano-electrodes) that can simultaneously measure the true and transformed waveforms, from the inside and outside of the cell, respectively. Herein, we propose to fully reconstruct high resolution potential waveforms of hippocampal neurons, merely from transformed signals by machine learning (ML) techniques. To this end, we will first develop a platform to collect a large dataset of true and transformed waveforms from spontaneously active hippocampal neurons using our nanoelectrodes. Next, we will use this dataset to train, test and validate an ML model on thousands of true and transformed potential pairs. We will also validate our ML model by comparing the predictions from our model with waveforms obtained from gold standard methods such as patch clamp. Based on promising preliminary data, we anticipate that our proposed method could reliably reconstruct high-resolution neuronal potential waveforms providing a new direction for in vitro and in vivo neuronal recording for neuroscientists.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 29, 2024
- Source ID
- FA95502310090
Entities
People
- Zeinab Jahed
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of California, San Diego