Robust Cross‐Batch SERS Quantification of Nitric Oxide Enabled by Au@Ag@Au Nanocubes and TabPFN Full‐Spectrum Learning

Quantitative analysis of gasotransmitters using surface-enhanced Raman scattering (SERS) in complex biological environments remains challenging. Reaction-based sensing mechanisms often induce coupled variations across multiple vibrational modes, weakening the robustness of conventional univariate calibration strategies. Moreover, batch-to-batch variations of plasmonic substrates and interference from biological matrices further limit the generalization capability of quantitative models. Here, we present a quantitative framework that integrates structurally controllable nanoprobes with full-spectrum regression based on in-context learning. Gold-core@silver-shell@gold-outer-shell nanocubes (Au@Ag@Au NCs) were synthesized to improve reproducibility of plasmonic responses across batches. Meanwhile, a Tabular Prior-Data Fitted Network (TabPFN) model was employed to capture nonlinear correlations among multidimensional spectral features without iterative retraining. Using nitric oxide (NO) as a representative gasotransmitter, the proposed strategy was first validated in artificial cerebrospinal fluid to assess robustness against matrix interference, and subsequently applied to monitor intracellular NO fluctuations in hydrogen peroxide-induced inflammatory cell models. Comparative experiments demonstrate that the TabPFN-based approach reduces the root mean square error (RMSE) by 66.69% in unseen matrix batches compared with conventional single-batch calibration methods. This work provides a practical solution for quantitative SERS analysis of gasotransmitters and highlights the potential of full-spectrum in-context learning for improving cross-batch robustness in complex biological sensing scenarios.