Machine learning classification of cellular states based on the impedance features derived from microfluidic...

Area of Science:

• Cell Biology
• Biophysics
• Computational Biology

Background:

• Mitosis is essential for cell division, and its disruption is a target for anti-cancer drugs.
• Drug-induced cell cycle arrest or apoptosis are key outcomes in cancer therapy.
• Label-free, high-throughput single-cell analysis is crucial for understanding drug effects.

Purpose of the Study:

• To apply machine learning (ML) with microfluidic impedance flow cytometry (IFC) for label-free classification of drug-treated tumor cells.
• To differentiate between various cell states, including G1/S, G2/M arrest, and apoptosis.
• To classify subpopulations of cells, including drug-resistant ones.

Main Methods:

• Utilized IFC to measure electrophysiological parameters of single cells.
• Extracted impedance amplitudes and phases as features for ML model training.
• Employed a deep neural network (DNN) model with optimized structures for cell type and drug-specific classification.

Main Results:

• Achieved high classification accuracies for H1650 and HeLa cells, differentiating between G1/S, G2/M arrest, and apoptosis states.
• Demonstrated high accuracy (approaching 100%) in classifying viable vs. non-viable cells and G2/M arrest vs. apoptosis in vinblastine-treated HeLa cells.
• Successfully classified three subpopulations (drug-insensitive, G2/M arrest, apoptosis) in taxol or vinblastine-treated cells.

Conclusions:

• ML combined with IFC provides a powerful, label-free method for high-throughput single-cell analysis of drug effects.
• The DNN model effectively classifies complex cell states induced by anti-cancer drugs.
• This approach has significant potential for drug discovery and personalized medicine by identifying drug responses and resistance mechanisms.