Abstract
Cellular reactive oxygen species (ROS), a key parameter involved in cell metabolism, signaling, and apoptosis, whose detection is necessary to achieve in a variety of biological processes. However, current ROS detection methods, including fluorescence, colorimetry, and electrochemical methods, are difficult to achieve in-situ non-invasive detection due to their reliance on invasive probes or destructive sampling. In this study, we propose an in-situ non-invasive ROS detection integrating the Raman spectrum and a bidirectional gated recurrent unit (Bi-GRU) deep learning model during HepG2 cell apoptosis. The Bi-GRU model leverages bidirectional gating mechanisms to capture long-term dependencies in Raman spectra while incorporating both forward and backward spectral information for enhanced feature extraction. After training with spectral data of HepG2 cells in various apoptotic states, the R(2) (coefficient of determination) of the Bi-GRU model reaches 0.8511, which outperforms that of traditional methods such as KNN (0.2607), PLS (0.4720), and RNN (0.7724). In the present study, we not only realized the in-situ and non-invasive cellular ROS detection but also expanded the application of artificial intelligence in the field of cellular medicine. Importantly, this will provide a new research idea for further understanding the physiological state of cells and the mechanism of drug action.