Quantum machine learning-based electrokinetic mining for the identification of nanoparticles and exosomes with minimal training data.

Synthetic and naturally occurring particles, such as nanoparticles (NPs) and exosomes; a type of extracellular vesicles (EVs), have garnered widespread attention across various fields, including biomaterials, oncology, and delivery systems for drugs and vaccines. Traditional methods for identifying NPs and EVs, such as transmission electron microscopy, are often prohibitively expensive and labor-intensive. As an alternative, the assessment of electrokinetic attributes such as zeta potential or electrophoretic mobility, conductance, and mean count rate, offers a more cost-effective, rapid, and reliable means of characterizing these particles. In this context, we introduce the first application of a quantum machine learning (QML)-based electrokinetic mining for the identification of green-synthesized iron- and cobalt-based NPs, as well as exosomes derived from human embryonic stem cells (hESC), human lung cancer (A549) cells, and colorectal cancer (CRC) cells, based solely on their electrokinetic attributes. Comparative analyses involving cross-validation, train-test splits, confusion matrices, and Receiver Operating Characteristic (ROC) curves revealed that classical ML techniques could accurately identify the types of NPs and EVs. Notably, QML demonstrated proficiency in differentiating between various NPs and EVs, including the distinction of EVs in the plasma of CRC patients versus those of healthy individuals. Furthermore, QML's application has been extended to the identification of NPs along with EVs in the plasma of CRC patients and experimental mice, achieving higher prediction performance even with a minimal training dataset, demonstrating that QML based electrokinetic mining could identify NPs or EVs with minimal training data, thereby facilitating novel clinical development in the realm of liquid biopsies.