Deep learning-assisted surface-enhanced Raman spectroscopy detection of intracellular reactive oxygen species

Realizing the intelligent analysis of the intracellular reactive oxygen species (ROS) is beneficial to quick diagnosis of diseases. Herein, surface-enhanced Raman spectroscopy (SERS) technology was combined with deep learning to establish a smart detection method of intracellular ROS based on neural network to improve the SERS analysis ability. Taking the simultaneous detection of peroxynitrite (ONOO^-) and hypochlorite (ClO^-) as the templates, 4-mercaptophenylboric acid (4-MPBA) and 2-mercapto-4-methoxyphenol (2-MP) molecules were modified on the AuNPs to prepare AuNP/4-MPBA/2-MP nanoprobes. The SERS spectra of AuNP/4-MPBA/2-MP nanoprobes before and after the specific response of ONOO^- and ClO^- were collected to construct a database, and the neural network model for extraction (ENN) and one-dimensional convolutional neural network model (1D-CNN) for quantification were built. The cosine similarity values of ENN model for ONOO^- and ClO^- correlation spectra were 0.997 and 0.995, respectively. In addition, the qualitative and quantitative results of the models were basically consistent with the experimental results. Moreover, the models can accurately extract the SERS response spectral information of ONOO^- and ClO^- and realize their preliminary prediction of concentration in living cells, which has great potential in the high-throughput smart processing and accurate analysis of large-scale complicated SERS data from biological system.