A Data-Driven and Interpretable Framework for Discovering New Analytical Detection Mechanisms in Surface-Enhanced Raman Scattering Sensors

Surface-enhanced Raman spectroscopy (SERS) offers rich molecular fingerprint information and holds great potential for quantitative chemical analysis in biosensing, diagnostics, and environmental monitoring. However, the development of quantitative SERS sensors remains largely empirical, relying on prior chemical intuition to select key spectral indicators, which limits efficiency and may overlook subtle yet informative features. Here, we present a Data-driven and Interpretable framework for Mechanism discovery in SERS (DIMS) sensing indicators without prior manual spectral indicator selection. This framework integrates high-throughput acquisition of Raman spectra at varying analyte concentrations, machine learning-based spectral classification, and interpretable analysis to identify spectral features that most strongly govern analyte quantification. The identified important features can be applied as quantitative or qualitative sensing indicators for the analyte after validation using chemical expertise. Using urea detection in artificial eccrine sweat as a model system, DIMS uncovered new spectral indicators and enabled reliable qualitative and quantitative sensing over a wide concentration range, extending down to 10 −8 M. DIMS provides an efficient strategy for discovering chemically meaningful sensing mechanisms and accelerating the development of high-performance SERS sensors. The proposed methodology is conceptually extendable to other spectroscopic techniques for interpretable feature discovery and molecular interaction analysis.