Uncovering details of the electrical properties of cells
Date
2022
Authors
Nejad, Jasmine E., author
Lear, Kevin L., advisor
Tobet, Stuart, committee member
McGrew, Ashley K., committee member
Simske, Steve, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
The electrical properties of cells have long been studied by scientists across many fields, yet there are still major gaps in our understanding of the intrinsic properties of many types of cells, such as parasite eggs, as well as the detailed electrical behavior of excitable cells, such as neurons. This work aims to provide insights into how these properties can be measured and how machine learning can be used to advance our understanding of these phenomena. The first part of this work discusses the development of a microfluidic impedance cytometer for the enumeration and classification of parasite eggs isolated from fecal samples. Current diagnostics in parasitology rely on the manual counting of eggs, cysts, and oocysts on microscope slides that have been isolated from fecal samples. These methods depend on trained technicians with expertise in the preparation of samples and detection of parasites on these slides, which increases cost and turnaround times for diagnosis. This leads many farmers and ranchers to opt to pool fecal samples from multiple animals to save time and labor. In cattle herds, resistance is often due to underdosing, which can be caused by treating all animals to an average weight or treating by the calendar instead of targeted deworming. This blanket use of anthelmintics, or anti-parasitic medication, is leading to concerns about anthelmintic resistance, which would cause major issues in the livestock industry, as well create unforeseen ecological imbalances. The developed microfluidic system provides a proof-of-concept for a microfluidic impedance cytometer capable of measuring the impedance of parasite eggs at multiple frequencies, simultaneously, as each of the eggs passes through a microfluidic channel past a sensing region. This region consists of parallel electrodes on the top and bottom of the channel, allowing for measurement of the voltage across the channel. When an egg passes through, the signal is interrupted, leaving a distinct profile of the electrical properties at each frequency over time. This system shows proof-of-concept of the impedance measurements at 500kHz and 10MHz and provides insights for further exploration of these properties, with the eventual use of machine learning algorithms for discrimination of parasite eggs from debris, and differentiation of parasite genera. The second part of this work discusses machine learning classification of neuronal subtypes based on features extracted from patch-clamp recordings from adult mice, using data acquired from publicly available databases. Classification of neuronal subtypes has been a continuously progressing area of neuroscience, building on advancements in our understanding of the morphology, physiology, and biochemistry of different neurons, and contributing to the accuracy and repeatability of action potential and neuronal circuit models. This work explores the use of k-nearest neighbors, support vector machine, decision tree, logistic regression, and naïve Bayes algorithms for classification of fast-spiking or regular-spiking neurons from the hippocampus or the primary somatosensory cortex. K-nearest neighbors shows the most accurate classification of these groups, using action potential width, amplitude, and onset potential as features (inputs into the algorithm), with the addition of a measure of rapidity (acceleration near action potential onset) showing major increases in classification accuracy. Of the three methods for measuring rapidity, inverse of the full width at half of the maximum of the second derivative of the membrane potential (V̈m) (IFWd2), inverse of the half width at half of the maximum of V̈m (IHWd2), and the slope of the phase plot (V̇m vs. Vm) near AP onset (phase slope), including the phase slope measure of rapidity increased the accuracy to nearly perfect (weighted f1-score > 0.9999). In addition, the use of phase slope and action potential width as the only features for classification produces measures of accuracy, weighted f1-scores, of >0.9996. The results show the value of rapidity in action potential dynamics, the distinct difference between rapidity in APs generated by hippocampal neurons relative to cortical neurons, and low standard deviations for rapidity values in cortical neurons (fast- and regular-spiking). These findings have potential implications for understanding the ion channel dynamics in action potential initiation and propagation, which can improve the modeling of action potentials and neuronal circuits.
Description
Rights Access
Subject
cytometry
neuron
rapidity
diagnostics
action potential
parasitology