Diagnostic test combining two technologies with machine learning could lead to new paradigm for at-home testing

A new diagnostic testing system jointly developed by the University of Chicago’s Pritzker School of Molecular Engineering (PME) and UCLA’s Samueli School of Engineering combines a powerful, sensitive transistor with a low-cost paper-based diagnostic test. Combined with machine learning, the system becomes a new type of biosensor that could eventually transform at-home testing and diagnostics.

Led by Professor Junhong Chen of the University of Chicago and Professor Aydogan Ozcan of UCLA, the research team combined a field-effect transistor (FET) – a device capable of detecting concentrations of biological molecules – with a paper-based analytical cartridge (the same type of technology used in at-home pregnancy and COVID tests).

This combination combines the high sensitivity of FETs with the low cost of paper-based cartridges. Combined with machine learning, the test measured cholesterol in a serum sample with an accuracy of over 97%, compared to results from the CLIA-certified clinical chemistry laboratory at the University of Chicago Medicine, led by KT Professor Jerry Yeo.

The research, published in ACS Nanowas conducted in collaboration with Ozcan’s team at UCLA, which specializes in paper-based detection systems and machine learning. The result is a proof of concept that could eventually be used to create inexpensive, highly accurate home diagnostic tests capable of measuring a variety of health and disease biomarkers.

“By addressing the limitations of each component and adding machine learning, we created a new testing platform that could diagnose diseases, detect biomarkers and monitor therapies at home,” said Hyun-June Jang, a postdoctoral researcher and co-senior author of the paper with Hyou-Arm Joung of UCLA.

Home diagnostic tests, such as pregnancy or COVID-19 tests, use paper-based testing technology to detect the presence of a target molecule. While these tests are simple and inexpensive, they are largely qualitative and inform the user whether or not the biomarker is present.

At the other end of the testing spectrum are FETs, originally designed for electronic devices. Today, they are also used as highly sensitive biosensors that can detect biomarkers in real time. Many believe that FETs are the future of biosensing, but their commercialization has been hampered by the specific requirements of testing conditions. In a highly complex matrix like blood, it can be difficult for FETs to detect a signal from an analyte.

Chen and Ozcan’s teams decided to combine the two technologies to create a new type of test system. The paper fluidic technology, and specifically its porous sensing membrane, reduced the need for a complex, controlled test environment normally required by FETs. It also provides a low-cost basis for the system, since each cartridge costs about 15 cents.

When the team integrated deep learning kinetic analysis, they improved the accuracy and precision of the test result within the FET.

“We’ve improved accuracy and created a device that costs less than fifty dollars,” Jang said. “And the FET can be reused with disposable cartridge tests.”

To test the system, the team programmed the device to measure cholesterol from anonymous human plasma samples. In 30 blinded tests, the system measured cholesterol with an accuracy of more than 97%, far exceeding the 10% total allowable error under CLIA guidelines.

The team also conducted a proof-of-concept experiment that showed the device could integrate immunoassays, which are widely used in the quantification of hormones, tumor markers and cardiac biomarkers.

“This is a much improved conventional diagnostic system, which will be important as home testing and diagnostics continue to gain popularity in the U.S. healthcare system,” Jang said.

The team will next develop an immunoassay system and hope to demonstrate that the system can detect multiple biomarkers from a single sample. “This technology has the potential to detect multiple biomarkers from a single drop of blood,” Jang said.

Other co-authors of the paper are Artem Goncharov, Anastasia Gant Kanegusuku, Clarence W. Chan, Kiang-Teck Jerry Yeo and Wen Zhuang.