Raman spectroscopy combined with machine learning algorithms for discrimination of cancer cell death pathways: Apoptosis and pyroptosis.

BACKGROUND AND

OBJECTIVES

Identifying drug-induced cancer cell death pathways is crucial for understanding the mechanisms of drug action. However, traditional biological experimental methods are limited by their time-consuming nature and high cost. This aim of this study is to develop classification models that can predict apoptosis and pyroptosis in cancer cells, combining Raman spectroscopy with machine learning algorithms.

METHODS

Cisplatin and dronedarone, two drugs known to induce distinct cell death pathways, were selected to validate the predictive performance of the constructed models. Additionally, Gradient-weighted Class Activation Mapping (Grad-CAM) visualization method was employed to identify key feature peaks. Model predictions were further validated through cellular morphological observations and Western blot analysis.

RESULTS

Both algorithms achieved prediction accuracies exceeding 90% for apoptosis and pyroptosis. Predictions of cell death pathways induced by cisplatin and dronedarone were concentrated on apoptosis and pyroptosis, respectively. Visualization results using gradient-weighted class activation mapping showed that apoptosis is associated with characteristic peaks of phenylalanine, while pyroptosis is associated with characteristic peaks of tyrosine. These results are consistent with existing research findings.

CONCLUSION

This study demonstrates that Raman spectroscopy combined with machine learning algorithms can accurately predict cancer cell death pathways, and that visualization techniques can be used to identify biomarkers associated with these pathways, providing reliable technical support for research into drug mechanisms of action.