NEW YORK – Researchers at the University of California, Davis have developed a mass spectroscopy-based assay that they say could potentially diagnose COVID-19 infections in 20 minutes. The team aims to pursue Emergency Use Authorization with its partner, SpectraPass, which will use the assay as an alternative to molecular and antigen screening tests.
The core enabling technology of the assay is a method called Machine Intelligence Learning Optimizer, or MILO, pioneered by researcher Nam Tran and his colleagues at UC Davis, that uses advanced computing methods to find patterns of peaks within messy mass spec data.
Up- and downregulation of different sets of proteins can be a biomarker of different infections, but there are potentially thousands of distinct proteins in nasal swab samples, Tran said in an interview. These proteins can in turn generate thousands of overlapping peaks in MALDI-TOF data. Tran and his team used MILO and showed that the machine learning program could be trained to sift through and detect relevant signals to accurately distinguish COVID-19-infected patients, as described in a recent Scientific Reports study.
The MILO machine learning method — which was originally developed for burn-related sepsis and organ transplant applications — was trained on samples from 107 symptomatic and asymptomatic patients who had COVID-19, as well as 92 samples from people with other non-COVID respiratory conditions and healthy controls.
According to the Sci Reports study, the MILO method employs a few different machine learning techniques. It has a collection of so-called "hyperparameter search tools" to crunch the MALDI-TOF peaks and help determine the optimal combination of algorithms or methods to attack the data, for example deep neural network, logistic regression, or random forest analyses. The MILO tool then combines the best data assessment methods to create thousands of unique machine learning pipelines, which in turn generate over 100,000 models that are statistically assessed to identify the best performing model.
The machine learning-based method finally generates a pipeline whereby patient sample MALDI-TOF data goes in and a readout of COVID-19 positive or negative comes out.
And rather than just picking up the proteins of the virus itself, the test seems to pick up the disease, according to Tran. In other words, the virus makes many different proteins and so does the host's immune system, and the many protein signatures MILO learned to recognize as "COVID-19 positive" likely include both.
With the data analysis piece in place, the MALDI-TOF peaks from patients' samples can now be quickly assessed.
Additionally, MALDI is a rapid technology with a reasonably high throughput, Tran added, and the team can use it to process 48 samples in 20 minutes, which could make it a useful screening test for businesses, airports, or schools. In contrast, the turnaround time to report high-throughput PCR tests can be days, while rapid antigen tests typically only test one person at a time.
"This technology is a nice middle ground," Tran said.
In general, MALDI-TOF methods are routinely used in microbiology labs as a way to identify pathogens grown in culture. Potential advantages of MALDI-based diagnostics include speed, different supply chain needs compared to other diagnostics, and numerous clinical instruments already installed.
But there have been few COVID-19 assays that use the method to date, Tran said, and the technology has apparently struggled to demonstrate utility during the pandemic.
One COVID-19 test that employs MALDI is from Agena Bioscience, which obtained EUA for a test that incorporates MALDI-TOF on its MassArray system in October. MassArray is also used with authorized tests from National Jewish Health Advanced Diagnostics Laboratory and Ethos Laboratories, and Agena recently partnered with the latter for a variant identification test.
The Agena test analyzes pieces of PCR-amplified nucleic acids using MALDI-TOF to make detection faster and more efficient. "But it is not a direct-to-MALDI technique," Tran said, adding that his team also considered this route but reasoned that it would still depend on a reliable supply of PCR reagents.
Other research teams have also been working on mass spec-based COVID-19 diagnostics. For example, a team at the University of Virginia School of Medicine is using a deep learning software tool called Prosit to generate spectral libraries to the SARS-CoV-2 proteome, while a team based in Chile developed a method for direct detection of the SARS-Cov-2 in nasal swabs. The latter method differs from the UC Davis method in that it tests from specialized transport media, rather than directly from patient swabs preserved in saline.
"I believe we are presently the only ones who have done direct swab-to-MALDI detection of proteins from a nasal sample," Tran said. He and his colleagues are using its instruments and assay for research use only to generate the data for larger clinical trials, Tran said, but hope their method could eventually receive EUA from the FDA with the help of SpectraPass.
Cofounded by Maury Gallagher, chairman and CEO of Allegiant Airlines, Las Vegas-based SpectraPass facilitates the authentication of customers' health, particularly airline travelers, and helps reopen businesses.
Despite increasing vaccination rates, Gallagher said in an email that "COVID testing doesn’t seem to be going away anytime soon." This is particularly true in the travel industry, where negative tests are required to enter some countries.
"Given our test's ability to scale quickly, we see it as a very efficient test to use for admittance into a country, and also a potentially more affordable option than test solutions currently on the market," Gallagher also said.
The assay could also be used to screen people prior to entering events, like concerts, conventions, and sporting events, or anywhere that social distancing is not feasible. For this rapid approach, MALDI-TOF instruments could be installed close to the customers who will be tested, such as in the airport or at the stadium.
Although the test originated with the COVID-19 pandemic, "we have demonstrated that mass spectrometry can be utilized to test for viruses, not just coronavirus," Gallagher said. "We don’t intend to stop with COVID."
Gallagher said he believes the partners have built a platform that can test for different types of viruses in a cost-effective, fast, and highly accurate manner.
The method scales well, as well. "My hope is that we would be able to use this system to detect and mitigate future pandemics, to ensure we never see another full closure of business and personal interaction as we’ve all endured with COVID," Gallagher said.
PCR-based host-response tests — from companies such as Immunexpress, Predigen, Inflammatix, or SpeeDx — tend to target a handful of biomarkers and can be used to distinguish viral and bacterial infections or to diagnose sepsis, and they have also been developed to predict severity of COVID-19 infections by detecting immune system activation.
A host-response MALDI approach using machine learning can generate a biosignature that is more complex, incorporating many peaks. And, it is unlikely to be as impacted by viral variants, the way PCR tests can be, Tran said. It could also theoretically be used to detect the known immune system activation signatures that predict severe COVID-19, or MILO could be trained on samples from severe cases to find its own predictive signatures.
Tran said his collaborators have had conversations with the FDA to pursue EUA, but also said that the published study could inspire others to use the assay as a lab-developed test. The UC Davis team uses a Shimadzu 8020 MALDI-TOF-MS analyzer but the assay should be instrument-agnostic, he said.
And, while there is no mass spec EUA template to work from at this stage, the team is in ongoing discussions with the FDA for guidance. "Using MALDI-TOF with machine learning and cloud-based technology [for diagnostics] ... this is all new," Tran said.
He said the team will begin to look at other respiratory infections besides COVID-19, and could adapt its MALDI-based test readily with the MILO technology.
"The machine learning allows us to see patterns that the human eyes may not be able to discern," he said.