W–O–Ti Bridge Modulated MXene Heterostructures and Hydrogel Electrolytes Enable Asymmetric Machine-Learning Decoupling for Thermally Robust Molecular Quantification

Abstract Ambient temperature fluctuations impose a fundamental barrier to the field deployment of high-precision chemical sensors, as nonlinear thermal drift often overshadows subtle concentration-dependent signals. Here, we report a materials-to-algorithm paradigm that achieves intrinsic thermal robustness through interfacial engineering and asymmetric algorithmic compensation. We synthesize a W18O49-decorated MXene (MXene-W) cathode featuring strong W–O–Ti electronic coupling, which modulates the local electron density to stabilize redox kinetics against thermal perturbation. This is integrated with a carrageenan-polyacrylamide (CG-PAM) quasi-solid hydrogel electrolyte that ensures stable ion transport across a broad temperature window. Mechanistically, we uncover a critical asymmetry: bioluminescent transduction exhibits a pronounced nonlinear temperature dependence (combined Temp/Temp2 contribution of 0.35), surpassing the signal’s intrinsic contribution, whereas electrochemical transduction is less temperature-sensitive. To resolve this, parallel response surface methodology streams feed into ensemble learners, effectively decoupling temperature-signal-concentration manifolds. The optimized platform achieves significant sensitivity and predictive accuracy (R2 = 0.9982). By deploying the inference pipeline as an offline application, we demonstrate cloud-free, on-site quantification of trace molecular analytes in complex matrices. This work provides a generalizable strategy for bridging nanoscale interfacial control with edge-intelligent computation to mitigate environmental noise in decentralized sensing networks.