Machine Learning-Based Detection of Anaplasma spp. Using Dielectric Properties of Host Cells.

Tick-borne bacterial infections such as those caused by Anaplasma phagocytophilum are difficult to diagnose, particularly at early stages, because they rely on specialized laboratory tests. Dielectric-based measurements provide a label-free way to probe cellular state and may offer useful information. In this study, we explored whether A. phagocytophilum-infected and uninfected HL-60 human promyelocytic leukemia cells can be distinguished using dielectric measurements collected under controlled in vitro conditions. Measurements were performed at two medium conductivities (100 and 300 µS/cm), and three dielectric properties were examined: cytoplasmic conductivity, specific membrane conductance, and specific membrane capacitance. We used exploratory analysis, statistical tests, and interpretable linear classifiers, with performance evaluated using leave-one-out cross-validation. Membrane-related properties, especially specific membrane conductance, showed the clearest separation between infection states and were consistently the most informative across models. Both classification performance and effect sizes were stronger at the higher medium conductivity condition. These results are encouraging, but because the sample size was small and consisted only of technical replicates, they should be interpreted as preliminary measurement-level evidence. This study is intended as a pilot and supports further work with larger datasets, additional cell types, and broader experimental conditions.