Machine learning-based cytokine microarray digital immunoassay analysis.

Area of Science:

• Biomedical Engineering
• Analytical Chemistry
• Critical Care Medicine

Background:

• Serial protein biomarker measurement is crucial for critical care but faces technological challenges in sensitivity, multiplexity, and speed.
• Existing assays struggle to provide rapid, accurate, and high-capacity multiplexed detection of low-concentration biomarkers.

Purpose of the Study:

• To develop a rapid, accurate, and highly multiplexed microfluidic digital immunoassay for near-patient bedside use.
• To overcome the limitations of current assays in sensitivity, speed, and multiplexing capabilities for critical care applications.

Main Methods:

• Development of a microfluidic digital immunoassay incorporating machine learning-based autonomous image analysis.
• Utilized a spatial-spectral microfluidic encoding scheme and a convolutional neural network (CNN) for single-molecule protein digital counting.
• Achieved 12-plexed biomarker detection in <15 μL sample volume with <5 pg/mL sensitivity and 5-min incubation.

Main Results:

• The assay demonstrated high sensitivity (<5 pg/mL) and multiplexity (12-plex) with a rapid turnaround time (<40 min from sampling to result).
• Machine learning-based image analysis significantly reduced errors in high-capacity multiplexing at low protein concentrations.
• Successfully profiled 12 serum cytokines in patients undergoing CAR-T cell therapy.

Conclusions:

• The developed microfluidic digital immunoassay offers a powerful tool for near-real-time biomarker profiling in critical care.
• This technology has potential applications in monitoring critically ill patients, including those with cytokine storm syndrome during COVID-19.
• The assay represents a significant advancement in transforming critical care through rapid, sensitive, and multiplexed biomarker diagnostics.