Researchers have developed a new methodology that uses artificial intelligence tools to identify and count target viruses more efficiently than previous techniques. The new approach can be used in applications such as pharmaceutical biomanufacturing.
“Many gene therapies rely on viral vectors, which are viruses that are engineered to deliver a genetic payload that has therapeutic properties,” says Michael Daniele, a professor of electrical and computer engineering at North Carolina State University and senior author of a paper on the work published in the IEEE Sensors Journal.
Right now, if biomanufacturers want to measure the amounts of viral vectors being produced in a specific process, they usually run a multi-step ELISA kit that tags the viral vectors in a sample with something that can then be detected by a sensor. This tends to be fairly time consuming and is either labor-intensive or makes use of expensive automated systems.
Electrochemical sensing meets machine learning
“We wanted to see if there was an electrochemical way of sensing these viral vectors, which wouldn’t require you to ‘tag’ the vectors in order to identify and count them,” says Daniele. “We found that you can detect these viral vectors using electrochemical techniques, but machine learning is needed to separate the signal that tells us about the viral vector from ‘noise’ caused by pH- and background variation.” Daniele is also a professor in the Lampe Joint Department of Biomedical Engineering at NC State and the University of North Carolina at Chapel Hill.
To address this challenge, the researchers made use of electrochemical impedance spectroscopy (EIS), which relies on a simple electrode substrate. The surface of the substrate is chemically functionalized to bind with the targeted viral vectors. As the vectors are bound to the surface of the functionalized electrode, the electrical signal changes. The change in the features of the electrical signal can then be used to estimate how many viral vectors are in the sample.
“That’s great, in theory,” says Daniele. “However, in reality, you often get other objects or molecules binding to the surface of the substrate, creating a lot of ‘noise’ that has historically made it difficult to get accurate readings using EIS without introducing a lot of additional steps into the process—which effectively make this process no more efficient than any other analytical technique for measuring viral vectors.”
Training AI to cut through noise
This is where AI comes in.
The researchers trained six widely used machine-learning models to quantify viral vectors from EIS data, using validated samples. They then ran a series of tests to evaluate how well each model performed at accurately counting viral vectors in dozens of samples. The samples varied in both the amount of viral vector present and in terms of the sample’s pH value, which reflects real-world sampling conditions and can also influence the electrical signal.
“We found that the models could easily determine the background pH values of each sample, which improved model performance at measuring the viral vectors,” says Daniele. “And the results are very promising.
“Basically, using an AI model in conjunction with EIS testing allows you to run samples very quickly and get fairly accurate viral-vector readings across a wide range of titers—it can look at viral vector concentrations across five orders of magnitude. If there’s a lot of virus present, it works well; if there’s very little virus present, it works well.
Complementing ELISA in biomanufacturing
“To be clear, ELISA is more precise—but it is a more time-consuming process, and it only works within a narrow range of titers,” says Daniele. “So, if there is a limited amount of virus in the sample, ELISA works well. But if there’s a lot of virus, you’d need to dilute the sample in order to get an accurate reading—which takes even more time.
“Ultimately, we think our new approach serves as an extremely useful tool for complementing ELISA in a wide range of contexts—particularly for biomanufacturing,” says Daniele.
“Basically, it allows you to quickly get readings at various points throughout the manufacturing process,” says Stefano Menegatti, co-author of the study and a professor of chemical and biomolecular engineering at NC State. “High-speed and rapid sensing is not just for the sake of speed alone, it allows us to implement rapid corrective measures as needed in the processes that produce and purify viral vectors.
“Another reason we’re excited about this work is that it ties in with another recent paper we did on how AI can be used to improve the purification processes used to isolate viral vectors at the end of the biomanufacturing process. Taken together, these papers highlight the potential that machine learning has to improve processes in biopharmaceutical manufacturing.”