Machine learning meets cytometry for anti-cancer drug performance analysis

A common class of anti-cancer drugs works by interfering with mitosis, the process of cellular division. Cancer cells affected by these drugs halt this process at a specific phase, preventing them from proliferating. Eventually, these cells die, shrinking the tumor.

Evaluating the effectiveness of these types of drugs requires analyzing the targeted tumor cells and determining how many are killed or frozen at the desired mitosis stage. Wei et al. detailed a method combining machine learning and impedance flow cytometry to classify the state of individual cells.

Traditionally, analyzing cell samples is done using biochemical measurement kits, which report an average viability across the cell sample. But these tools cannot differentiate states at the individual cell level.

“Monitoring the co-existing cellular states reveals the cell heterogeneity regarding the drug sensitivity at single-cell resolution,” said author Xiaoxing Xing.

Fluorescence flow cytometry can provide this information, but this method requires expensive reagents and, in some cases, prevents the cells from being cultured for additional analysis. In contrast, the authors’ approach using impedance flow cytometry along with machine learning can differentiate between cellular states non-destructively.

When testing their method, the team could accurately classify the states of both HeLa and H1650 cells.

“For HeLa cells, the classification accuracy is among the highest for machine learning-based cell-state classification on drug-treated tumor cells,” said Xing. “However, for H1650 cells, there is still room to improve.”

The researchers plan to explore additional machine learning models and increase the number of sampling frequencies over which they acquire impedance data to further improve the performance of their method.

Source: “Machine learning classification of cellular states based on the impedance features derived from microfluidic single-cell impedance flow cytometry,” by Jian Wei, Wenbing Gao, Xinlong Yang, Zhuotong Yu, Fei Su, Chengwu Han, and Xiaoxing Xing, Biomicrofluidics (2024). The article can be accessed at https://doi.org/10.1063/5.0181287 .