Using machine learning to determine the differences in the dielectric properties between samples that have been treated with antibiotics and untreated samples let the researchers identify the susceptibility of the bacteria to a given antibiotic in as little as two hours after treatment.
Antibiotic-resistant bacteria are becoming more common, and the consequences for our healthcare will be profound. Routine surgical procedures could become fraught with risk and simple infections could progress into more serious issues without any available drugs to combat them. However, thankfully, we are not at this stage quite yet, and we still have time to prolong the utility of our existing antibiotic arsenal. In this context, using the correct antibiotic drug is important to achieve the desired treatment outcome, and also to reduce the likelihood that resistance develops further
Getting a readout on the antibiotic susceptibility of the bacteria causing problems for a particular patient is best completed quickly so that the patient can avail of the correct treatment as soon as possible. However, current approaches to achieve this can take too long. “Oftentimes susceptibility results are needed much faster than conventional tests can deliver them,” said Yaxiaer Yalikun, a researcher involved in the study. “To address this, we developed a technology that can meet this need.”
The new technology involves using impedance cytometry to measure the dielectric properties of the bacteria. These properties will change quite quickly on contact with an antibiotic that the bacteria are susceptible to. The researchers split the bacterial sample in two, and treat one of these samples with an antibiotic before separately analyzing treated bacteria and untreated controls. Then, a machine learning algorithm learns the characteristics of the untreated bacteria, and determines if there is anything different with the treated cells.
“Although there was a misidentification error of less than 10% in our work, there was a clear discrimination between susceptible and resistant cells within 2 hours of antibiotic treatment,” said Yoichiroh Hosokawa, another researcher involved in the study.
Reference:
Tao Tang, Xun Liu, Yapeng Yuan, Ryota Kiya, Tianlong Zhang, Yang Yang, Shiro Suetsugu, Yoichi Yamazaki, Nobutoshi Ota, Koki Yamamoto, Hironari Kamikubo, Yo Tanaka, Ming Li, Yoichiroh Hosokawa, Yaxiaer Yalikun. Machine learning-based impedance system for real-time recognition of antibiotic-susceptible bacteria with parallel cytometry. Sensors and Actuators B: Chemical. Volume 374. 2023.
Abstract:
Impedance cytometry has enabled label-free and fast antibiotic susceptibility testing of bacterial single cells. Here, a machine learning-based impedance system is provided to score the phenotypic response of bacterial single cells to antibiotic treatment, with a high throughput of more than one thousand cells per min. In contrast to other impedance systems, an online training method on reference particles is provided, as the parallel impedance cytometry can distinguish reference particles from target particles, and label reference and target particles as the training and test set, respectively, in real time. Experiments with polystyrene beads of two different sizes (3 and 4.5 µm) confirm the functionality and stability of the system. Additionally, antibiotic-treated Escherichia coli cells are measured every two hours during the six-hour drug treatment. All results successfully show the capability of real-time characterizing the change in dielectric properties of individual cells, recognizing single susceptible cells, as well as analyzing the proportion of susceptible cells within heterogeneous populations in real time. As the intelligent impedance system can perform all impedance-based characterization and recognition of particles in real time, it can free operators from the post-processing and data interpretation.
