Researchers led by UCLA professor Aydogan Ozcan developed a deep learning-enabled biosensor for multiplexed, point-of-care (POC) testing of disease biomarkers. POC biosensors provide remote and resource-limited communities with an economical, practical alternative to centralized laboratory testing. The UCLA-developed POC sensor includes a paper-based fluorescence vertical flow assay to simultaneously detect three biomarkers of acute coronary syndrome from human serum samples. The vertical flow assay is processed by a low-cost mobile reader, which quantifies the target biomarkers through trained neural networks. According to the researchers, the competitive performance of the multiplexed computational fluorescence vertical flow assay, along with its inexpensive, paper-based design and hand-held footprint, give the POC sensor promise as a platform to expand access to diagnostics in resource-limited settings. “Compared to a commonly used linear calibration method, our deep learning-based analysis benefits from the function approximation power of neural networks to learn nontrivial relationships between the multiplexed fluorescence signals from the paper-based sensor and the underlying analyte concentrations in serum,” researcher Artem Goncharov said. “As a result, we have accurate quantitative measurements for all three biomarkers of interest, despite the background noise present in clinical serum samples.” Unlike lateral flow assays, which are the most common type of POC test, assays using the vertical flow of samples through stacked paper layers enable the arrangement of sensing regions in a 2D or 3D array and can achieve multiplexing with tens or even hundreds of independent testing channels represented by different affinity capture molecules. The vertical flow design of the POC sensor from the UCLA researchers has room for multiple test regions, with up to 100 individual test spots within a single disposable cartridge. “This design essentially allows us to integrate tens of different POC sensors into a single cassette and perform multiplexed diagnostics tests in parallel with the same low-cost paper-based sensor,” Ozcan said. Researchers at UCLA developed a deep learning-enabled, multiplexed point-of-care sensor that uses a paper-based fluorescence vertical flow assay. Courtesy of the Ozcan Lab at UCLA. The researchers used conjugated polymer nanoparticles (CPNs) — fluorescent labels with tunable emission and excitation properties and with minimally overlapping excitation and emission peaks — to design the fluorescence vertical flow assay. The CPNs have 480-nm excitation and 610-nm emission peaks, which helped the team reduce the strong autofluorescence background from the paper substrate. The excitation energy transfer in CPNs takes place across the whole backbone, catalyzing an amplified emission that is higher than quantum dots (QDs). CPNs are also more stable on porous paper layers, with less photobleaching, and are larger than QDs, leading to improved luminescence. Using human serum samples to quantify three cardiac biomarkers — myoglobin, creatine kinase-MB, and heart-type fatty acid binding protein — the researchers validated the fluorescence vertical flow assay platform. The assay achieved less than 0.52 ng/mL−1 limit-of-detection for all three biomarkers, with minimal cross-reactivity. Biomarker concentration quantification, using the assay coupled to neural network-based inference, was blindly tested using 46 individually activated cartridges and human serum samples. The results showed a high correlation between the fluorescence vertical flow assay and the ground truth concentrations obtained through standard laboratory benchtop testing, with a greater than 0.9 linearity and a less than 15% coefficient of variation found for all three biomarkers. The simple-to-operate POC sensor, the researchers said, involves only three injection steps performed through a single loading inlet. The steps can be executed by a minimally trained technician using a custom operation kit. The assay uses 50 µL of serum sample per patient and takes under 15 minutes to complete, which is on the same scale as, for example, COVID-19 rapid antigen tests that take between 15 and 30 minutes. The research was published in Small (www.doi.org/10.1002/smll.202300617).