West LA Times

Lyme disease, transmitted to humans through tick bites, presents significant diagnostic challenges. Initial symptoms include headaches, pain, and fatigue but can progress into long-term inflammatory conditions affecting joints, nerves, the brain, and the heart.

Currently, the gold standard for Lyme disease diagnosis is a two-part lab test that can take up to two weeks for results and often misses early-stage cases. However, new testing technology under development at UCLA promises quicker and more accurate detection. Utilizing artificial intelligence in a format similar to home COVID-19 tests, this new method has shown accuracy in identifying Lyme disease within 20 minutes with a single test.

Lyme disease affects over 600,000 people annually in the U.S., caused by a spiral bacterium spread through deer tick bites. The Centers for Disease Control and Prevention recommend a two-tiered testing regimen performed at centralized labs with results taking one to two weeks. According to the Bay Area Lyme Foundation, current testing methods miss seven out of ten early-stage cases.

Researchers from the California NanoSystems Institute at UCLA have developed a portable reader using AI that interprets results within 20 minutes. In a study published in Nature Communications, this AI-enhanced test was found as reliable as traditional methods requiring two tests.

“A lot of folks find out they have Lyme disease well after the point at which they could have been treated very easily,” said co-corresponding author Dino Di Carlo from UCLA Samueli School of Engineering. “If we can measure rapidly... then testing can be done more routinely.”

The new test involves dropping a blood serum sample into a cartridge followed by buffer fluid flowing vertically through layers of sponge-like paper loaded with lab-made peptides detecting antibodies formed in response to Lyme bacteria. The pattern formed on the paper is captured digitally and analyzed by an AI algorithm providing results. Each test paper costs $3 and uses an adapted $200 off-the-shelf smartphone reader.

Co-corresponding author Aydogan Ozcan emphasized the importance of quantifying multiple indicators from a single sample: “In Lyme disease... we needed AI to make sense of such a complex signal.” The researchers trained their algorithm using samples provided by the Bay Area Lyme Foundation’s Lyme Disease Biobank achieving 95.5% sensitivity and 100% specificity.

Ozcan noted: “AI is only as good as the data,” highlighting collaboration with the CDC for well-characterized samples crucial for their AI's learning process.

Synthetic peptides from Biopeptides Corp., were critical for this technology due to their specificity and stability compared to whole proteins used in some lab tests.

Di Carlo added: “We want to focus on responses that are very specific to Lyme disease... And at the same time... return fewer diagnostic errors.”

The researchers aim to bring this technology into clinics within a few years while seeking partners for scaling up production. They are also working on adapting it for whole blood samples and developing an independent AI sample reader.

The study’s co-first authors are Rajesh Ghosh and Hyou-Arm Joung from UCLA along with several other contributors from UCLA, New York Medical College, Biopeptides Corp., and Elizabeth Horn from the Lyme Disease Biobank. Support came from the National Institutes of Health and National Science Foundation.