Machine Learning Boosts Biochemical Sensor Performance by "Two Orders of Magnitude"

A team of scientists at the Pennsylvania State University has developed a way to boost the performance of biosensors by combining individual electrochemical sensing devices with machine learning — potentially opening up the path to more accurate non-invasive medical testing based on saliva or sweat.

"We developed a new approach to improve the performance of electrochemical biosensors by combining machine learning with multimodal measurement," explains Aida Ebrahimi, assistant professor of biomedical engineering, of the team's work. "Using our optimized machine learning architecture, we could detect biomolecules in amounts 100 times lower than what conventional sensing methods can do."

The team's work focused on the use of a single-material sensor, which reacts to the presence of tyrosine and uric acid and sweat and saliva. Readings from the sensor were used to train machine learning models, with the team experimenting with different approaches before settling on an optimized multi-modal approach that was found to boost the limit of detection by two orders of magnitude compared to conventional approaches to the same problem.

"Our methodology successfully used one material to differentiate and distinguish four neurochemicals that are important in diseases like Parkinson's and Alzheimer's," says Ebrahimi. "While this preliminary data is promising, we must work further to be able to detect the lower levels of these neurochemicals in biological samples such as saliva."

To prove the concept still further, the team developed a flexible wearable sensor patch based on the same sensor hardware and optimized machine learning model — successfully detecting both uric acid and tyrosine in the user's sweat over a wide range of concentrations. Further work will, the researchers hope, result in a low-cost handheld device for diagnoses in the field.

"The machine learning-powered electrochemical diagnostic approach presented in this paper may find broader application in multiplexed biochemical sensing," adds Vinay Kammarchedu, first author on the paper, of the technology's possibility. "For example, this method can be extended to a variety of other molecules, including food and water toxins, drugs and neurochemicals that are challenging to detect simultaneously using conventional electrochemical methods."

The team's work has been published in the journal Analytica Chimica Acta under closed-access terms, with a preprint available on ECSarXiv.