Skip to main content
Premium Trial:

Request an Annual Quote

Israeli Researchers Develop 40-Minute Method for Pathogen Detection


NEW YORK – Researchers in Israel have developed a method to detect bacteria and antibiotic resistance directly from urine, leveraging infrared spectroscopy technology and machine learning.

Developed by scientists at Ben-Gurion University, Shamoon College of Engineering, and Afeka Tel Aviv Academic College of Engineering, the method takes about 30 to 40 minutes to return results defining what kind of bacteria is present, as well as the bacteria's susceptibility to different antibiotics.

BGN Technologies, the technology transfer company of BGU, has licensed the technology and has filed for patent protection worldwide, including with the US Patent and Trademark Office. According to Galit Mazooz-Perlmuter, BGN's senior VP of business development, biopharma, the firm is now "looking for a strategic partner to advance and bring the technology to market."

The company declined to comment further about its commercialization strategy, saying it would be defined by the strategic partner.

Antibiotic resistance has been a major concern for years with no signs of easing. As a result, groups such as the Combating Antibiotic-Resistant Bacteria Biopharmaceutical Accelerator and the US National Institutes of Health have significantly invested in new ways to detect bacterial resistance and susceptibility to antibiotics, emphasizing the demand for novel technologies and quicker results.

To date, most methods for detecting antibiotic susceptibility and resistance have relied largely on culturing bacteria from a sample, which can take days to deliver a result. But Mahmoud Huleihel, head of the Shraga Segal Department of Microbiology, Immunology, and Genetics at BGU and one of the new method's inventors, said that the main thing clinicians want with infection detection is speed.

Instead of sending samples to the microbiology laboratory and waiting "at least 48 hours" to get results, the method he and his colleagues have developed aims to shorten the time-to-result, allowing for faster treatment of patients.

When results take days, physicians end up starting patients on antibiotics that may not be effective and could expose bacteria to an antibiotic that allows it to develop mutations, Huleihel said. Further, giving the patient the wrong treatment can also lead to further complications.

The method he and his colleagues have developed is based on infrared spectroscopy combined with machine learning to detect minor changes and mutations in cells. Infrared spectroscopy has been used for nearly four decades by researchers to identify various samples, including cancer cells, fungi, and bacteria at the strain level.

Two studies published by the researchers in 2017 provided proof of concept for the use of infrared spectroscopy and their machine learning algorithm with urine samples after 24 hours of culture, comparing the new method to traditional methods using mass spectrometry and disk diffusion to determine the bacteria species type and susceptibility to antibiotics.

A further study, published in 2019, built on the findings of the previous studies and increased the sensitivity and specificity of the method.

In one paper published in 2017 in the Royal Society of Chemistry's Analyst, the researchers noted that many antibiotic resistance mutations in bacteria "have developed as a result of repeated unnecessary use of these antibiotics, particularly for a treatment time that is insufficient to eliminate all of the bacteria."

In another 2017 paper published in Analytical Chemistry, the Israeli group wrote that "the potential of IR spectroscopy to identify chemical components by analysis of their vibrational spectra fingerprints can be of great value."

In the Analyst study, which was retrospective, sensitivity of the algorithm the group developed for a variety of antibiotics ranged from 64 percent to 84 percent, while in the Analytical Chemistry study, sensitivity ranged from 72 percent to 88 percent.

Building on the previous studies, in the 2019 study, when detecting extended-spectrum β-lactamase-producing bacteria from bacterial colonies after the first 24 hours of culture, the method had a 97 percent success rate, 99 percent sensitivity, and 94 percent specificity in 837 samples.

In the most recent development, the researchers said that their method was able to detect the genus and species of bacteria, as well as its resistance to antibiotics, directly from urine samples without the need for any culture, Huleihel said.

For the direct, culture-free method from urine, researchers used infrared spectroscopy with purified urine samples, hitting the sample with different wavelengths to get a "fingerprinting of the bacteria," Huleihel said. Purifying the bacteria from the urine sample via centrifugation and several washings takes approximately 15 minutes, and putting a small amount of the sample on the appropriate slide and drying it takes about five minutes, he said.

The current version of the method has shown approximately 80 percent sensitivity, which Huleihel called "highly similar" to the results obtained after the first culture.

What makes the method unique, Huleihel noted, is the ability to detect bacteria and identify susceptibility directly from a urine sample within 30 to 40 minutes after receipt of the sample.

According to Huleihel, various components of the bacteria cells absorb part of the infrared waves at different wavelengths, allowing the researchers to detect minor changes or mutations. These changes can't be seen with the naked eye, he said, so machine learning is necessary to detect them, adding the spectroscopy measuring spectra in the bacteria takes "seconds."

Once the bacteria fingerprints are taken, the spectroscopy results are analyzed by the machine learning algorithm developed by the researchers, which has been trained to detect those microscopic changes and determine the species of bacteria and whether it's resistant or sensitive to the examined antibiotics, Huleihel said.

The algorithm uses a framework of dimensionality-reduction followed by multidimensional classification. According to the Analyst paper, for a specific antibiotic the algorithm "was designed to distinguish between spectra found to be sensitive to the antibiotic from spectra that were found to be resistive," based on the gold standard of culture.

The algorithm can produce three potential outcomes for a sample: the bacteria is sensitive to a type of antibiotic; the bacteria is resistant to a specific antibiotic; or no outcome. For cases where the algorithm has high confidence, it returns a result of sensitive or resistant. In cases where there's low confidence, the algorithm can't detect sensitivity or resistance. The level of confidence is measured by the risk of misclassification, or disagreement with results from the full culture.

In a result with an outcome, further analysis by the algorithm can determine which antibiotic would be most effective among the available options.

Currently, the algorithm has been trained on 1,400 samples, but Huleihel said its accuracy will improve as it's tested on more samples. While the researchers have been preparing each sample for spectroscopic measurements, Huleihel said eventually it would be possible to prepare 100 samples simultaneously, including all of the purification steps of the bacteria from the urine, although using the method with multiple samples hasn't been tested yet.

Despite the emphasis on urine samples, the method could potentially be used with other sample types and tissues as well, Huleihel said. Urine samples were used because most infections arriving to medical centers are urinary tract infections caused by various bacteria, and "if the infection is in the urine system … it is everywhere," he added.

Huleihel noted that this method isn't for use in every lab, as the infrared spectroscopy instruments cost around $70,000, which may be too much for regular hospital laboratory to afford.

Gary Hastings, an associate professor of physics and astronomy at Georgia State University who has studied infrared spectroscopy, said that many groups in the field "use machine learning or neural networks to assess spectral differences."

But if the researchers can truly distinguish "a wide range of bacterial species … and assess antibiotic susceptibility in all strains, in the actual urine, then that would be pretty novel," he said.