|
Written by: Chris Shin '24 Edited by: Elaine Wang '24  Image 1: A diagram illustrating how individual viral particles pass through the nanopores via electrophoresis, allowing for the detection of drops in current. [1] One of the primary struggles the world faces today is a shortage of tests to detect the presence of the SARS-CoV-2 virus. Most recently, a group of researchers from Osaka, Japan, have proposed a new form of COVID-19 testing, one that is fundamentally based on machine learning, algorithms that improve itself without further intervention.[1] The algorithm developed has proven to detect single viruses and can distinguish the Coronavirus from other common respiratory viruses such as influenza and the adenovirus. The device for viral detection is rather straightforward; by drilling pores 300 nm wide into a silicon nitride membrane placed above a silicon wafer, the conduit allows for the flow of electric charges that will eventually be detected. By coating the conduit with an organic polymer, PDMS, viral samples can be readily placed into the nanopores. When voltages are applied to the wafer, ions pass through the nanopores via electrophoresis (where molecules travel under an electric field), allowing for a constant current to be detected. However, when viruses are also introduced into the nanopores, some ions are prevented from crossing from one end of the wafer to the other, leading to a decrease in measured current. Because each kind of virus induces decreases in current of different magnitudes, the researchers realized that the viruses could be distinguished by species based on size of the observed decrease in current. Taking into account the factors that could lead to these varying drops in current such as the area, the positioned angle, and the rotational inertia of the viral samples, the researchers created parameters for the machine to analyze and learn from. The machine used, called The Waikato Environment for Knowledge Analysis, can train classifiers (entities responsible for the “learning” part of machine learning) to identify common parameters across multiple samples of the same virus. In single trials of introducing voltages and viruses, the classifiers exhibited F-measure scores—the measure of the algorithm’s accuracy—of 71.3% to 87.3%. However, when signals were sent 25 times, the classifiers could identify a particular virus with an impressive 99% accuracy, “high enough for exploiting the method for the infection diagnosis”, the researchers wrote. This method of diagnosis not only has the potential for great accuracy but is also adaptable, as the classifiers may be taught to distinguish between old and new strains of the Coronavirus using the algorithm, a unique and rare feature among current testing techniques. Although no method of diagnosis can be perfect, this novel technique will revolutionize traditional testing methods as it is significantly less time-consuming and less technical, making it an approachable alternative for the public to use in the near future. Sources: Arima, A., Tsutsui, M., Yoshida, T., Tatematsu, K., Yamazaki, T., Yokota, K., Kuroda, S., Washio, T., Baba, Y., & Kawai, T. (2020). Digital Pathology Platform for Respiratory Tract Infection Diagnosis via Multiplex Single-Particle Detections. ACS Sensors, 5(11), 3398–3403. https://doi.org/10.1021/acssensors.0c01564
0 Comments
Leave a Reply. |
