Automating data classification for label-free point-of-care biosensing in real complex samples

Surface-based affinity biosensors present a promising avenue for point-of-care (POC) pathogen detection in real-world samples. While laboratory-based devices commonly employ various techniques to mitigate noise, signal drifts, fluidic artifacts, and other system imperfections, their simple cost-effective POC counterparts designed for field use frequently lack such capabilities. This paper addresses this gap by introducing a procedure for automatically classifying pathogen presence in unprocessed liquids from direct detection data measured by a simple POC quartz crystal microbalance sensor device. The procedure integrates classical analytical tools such as filtering, data selection, baseline de-drifting, and result calculation in tailored successive steps, considering the nature of the sensor signal and the challenges posed by real-world media. We show that the developed procedure exhibits exceptional robustness across different biosensing assays and complex real-world media. Through optimizing parameters using diverse datasets encompassing Escherichia coli O157:H7 (E. coli) and SARS-CoV-2 detection in various media including food-derived matrices and cell culture media, we achieved rates of successful detection as high as 80.8% and 90.9% for E. coli and SARS-CoV-2, respectively, without extensive machine learning. Furthermore, we analyse the sensitivity of the procedure to variations of input parameters and with examples discuss key factors influencing overall procedure accuracy. Our results suggest that this exceptionally robust method holds potential as a straightforward tool for automating sample classification in point-of-care diagnostics, underpinning its promising broader applicability.