Like a Fingerprint

In this Q&A, Swami shares his research methods and perspective on electrical engineers’ contributions to diagnostic tools for precision medicine.

Q. The nation’s COVID-19 response has relied in large part on testing. Even with vaccines’ availability, testing remains an important element in containing the spread of the disease. How might your research help doctors and public health officials track COVID-19 cases?

The most familiar test for COVID-19 is the nasal swab, also called a polymerase chain reaction (PCR) test. It detects genetic material from a specific organism, such as a virus. It then compares the genetic sequence of the nucleic acid—the RNA—inside the cells collected with known RNA sequences of COVID-19. The PCR test returns a straightforward yes-no response, whether you are infected or not.

The nasal swab and other tests of cells’ genetic makeup have their limitations, however. They cannot tell you how infective you are or how long you have been infected.

Our sensing approach overcomes these limitations. Specific to COVID-19, we would look at how the spike protein key opens the lock of a healthy cell. More broadly, we use a cell’s physical characteristics—technically speaking, its phenotype—such as its protein layer, its structure, and its secretions, to measure the number and type of shed virus particles that can bind to host cells. This correlates to the individual’s infectivity.

The exact quantity of the virus is very important to know, especially in situations in which underlying conditions like asthma accentuate COVID’s harm. These individuals will feel the effects of COVID-19 at a lower dose of the virus compared with individuals who are in full health.

We see our approach as a complement to rather than a replacement for COVID-19 PCR-type
tests. On the assumption that COVID-19 variants will be an ongoing public health concern akin to
the seasonal flu, we hope to offer more focused, second-level tests to determine whether a person’s viral load is at or above a threshold that makes the disease dangerous to them or others they
may contact.

Q. Your analogy to fingerprinting suggests that you need a huge database and machine learning to quickly analyze cells’ physical characteristics. Does that database exist today?

We need data, yes, but we envision a shared network of crowdsourced data repositories rather than a single, centralized database.

We recently published a paper, “Single-cell Microfluidic Impedance Cytometry: From Raw Signals to Cell Phenotypes Using Data Analytics,” selected as a cover article for Lab on a Chip. It presents data analytics for high-throughput phenotypic identification of cells, bacteria, and viruses. The journal is available through MedLine and reaches a wide audience of physicians, clinicians, and lab techs.

We partnered with Paolo Bisegna and Federica Caselli in the department of computer science at the University of Rome. They specialize in machine learning methods applied to the biophysical data that we generate in my lab. I should also mention that my postdoc Carlos Honrado was first author of the paper. We have another paper submitted in collaboration with the Air Force Research Lab that involved Yi Liu, another postdoctoral scientist in my research group, that delves further into the detection paradigm, specifically the selection of the receptor for binding the virus’ surface-layer protein.

We aim to educate the medical community about this new way to measure infectivity based on biomarkers.

Our paper in Lab on a Chip proposes a way to compare apples to apples when looking at two or more samples, that is, to normalize impedance data across measurements and labs. We also encourage doctors and labs to use our sensor in their own context, on a variety of viral and bacterial infections and chronic diseases.

In this way, we can develop a data system and architecture of biophysical models, share the data code, and enable others to write their own code. If we successfully promote this structure to quantify and classify data, we can more easily look for significant patterns such as immune reaction.

As more groups participate in using this data structure, we can relate biomarkers to particular diseases. This is a big leap, however, and it will take a sustained, long-term effort to make it happen. We need to identify the particular chemical and physical characteristics of each cell phenotype of interest and model how it interacts with another cell to understand how a disease-carrying cell affects the whole body. And we would need to repeat this process for each disease.