Machine learning-assisted plasmonic sensing using Au@Ag-Ti3C2Tx MXene/Au hybrid interfaces for high-efficiency and ultrasensitive microRNA detection.

Point-of-care nucleic acid testing requires both high sensitivity and rapid turnaround, yet conventional plasmonic biosensors often sacrifice speed for accuracy. Here, we present an integrated sensing strategy that combines a nanomaterial-enhanced plasmonic interface with intelligent time-series prediction to achieve minute-level and ultrasensitive microRNA detection. A uniform Au@Ag-Ti3C2Tx MXene layer is constructed on a gold (Au) -coated tilted fiber Bragg grating surface plasmon resonance (TFBG-SPR) sensor. The interface excites strongly coupled SPR-localized SPR hybrid plasmonic modes and efficient interfacial charge transfer between metal nanostructures and MXene. This cascade enhancement amplifies the electric field, enabling fM-level detection of microRNA-21 in noninvasive saliva matrices using a multi-channel platform. To accelerate quantification, a binding response curve reconstruction framework driven by a grey wolf optimizer-assisted CNN-LSTM model is established. The model accurately reconstructs the full response evolution without requiring equilibrium by learning temporal signatures from abbreviated early-stage spectra. This "short-to-long" strategy reduces the effective assay time to 4 min (≈73% reduction). In an independent validation dataset, the method exhibits robust signal recovery of 92-105%. It also achieves 100% quantitative accuracy on a single-order-of-magnitude-based interval quantification scheme. This dual-innovation system integrates a hybrid plasmonic enhancement interface with intelligent temporal response modeling. It provides a sensitive, efficient, and scalable platform for nucleic acid detection.