Label-free, machine-learning assisted impedance assessment of basophil activation

Abstract

Food allergies are increasing in prevalence, yet current diagnostic tests remain either inaccurate or potentially unsafe. The basophil activation test (BAT) has emerged as a promising functional assay, involving the ex vivo stimulation of whole blood with allergens and subsequent measurement of basophil activation. The BAT has been shown to have high diagnostic accuracy. However, clinical adoption of the BAT has been limited, in part due to logistical constraints, including the requirement for fresh blood and reliance on conventional flow cytometry analysis (FCA) of fluorescently labeled activation markers such as CD63 and CD203c. Despite recent advances in automation, FCA remains restricted to research and high-complexity clinical laboratories. Label-free, electronics-based sensing platforms have the potential to offer simpler operations and broaden the adoption of the BAT. In this study, we demonstrate the potential of label-free impedance flow cytometry (IFC) combined with machine learning for predicting basophil activation status. We measure impedance at six different frequencies (spanning 0.1 MHz to 24 MHz) in human basophils stimulated with different doses of stimulants from N = 15 anonymous donors. We compare IFC measurements with activation levels measured in FCA. By combining with machine learning models, we demonstrate strong correlation between IFC metrics and activation levels measured in FCA with Pearson correlation coefficient up to 0.89, and accurate classification of a positive or negative BAT result with a true positive rate (TPR) of 96% and a true negative rate (TNR) of 88%.