Rapid Identification\nof Drug Mechanisms with Deep\nLearning-Based Multichannel Surface-Enhanced Raman Spectroscopy

Rapid identification of drug mechanisms is vital to the\ndevelopment\nand effective use of chemotherapeutics. Herein, we develop a multichannel\nsurface-enhanced Raman scattering (SERS) sensor array and apply deep\nlearning approaches to realize the rapid identification of the mechanisms\nof various chemotherapeutic drugs. By implementing a series of self-assembled\nmonolayers (SAMs) with varied molecular characteristics to promote\nheterogeneous physicochemical interactions at the interfaces, the\nsensor can generate diversified SERS signatures for directly high-dimensionality\nfingerprinting drug-induced molecular changes in cells. We further\ntrain the convolutional neural network model on the multidimensional\nSAM-modulated SERS data set and achieve a discriminatory accuracy\ntoward 99%. We expect that such a platform will contribute to expanding\nthe toolbox for drug screening and characterization and facilitate\nthe drug development process.