Machine Learning Decoding of Iron-Doped Carbon Dots Dual-Mode Responses for Portable Smartphone Sensing and Real-Time Bioanalysis of Chiral Ascorbic Acid.

Achieving selective recognition and in situ monitoring of chiral isomers in complex biological environments remains a major analytical challenge. This study presents a portable sensing platform that combines machine learning with iron-doped carbon dots (FeCDs) for differential dual-mode detection and real-time bioimaging of chiral ascorbic acid (L-AA and D-AA). The FeCDs exhibit distinct chirality-dependent fluorescence responses: L-AA triggers a pronounced "red-to-cyan" fluorescence blue-shift by inducing the reduction of Fe3+ to Fe2+ and its subsequent dissociation from the carbon skeleton surface, thereby blocking ligand-to-metal charge transfer. In contrast, D-AA leads to a "red-to-colorless" fluorescence quenching predominantly via weak interactions governed by steric hindrance, following a photoinduced electron transfer pathway. DFT and IGMH analyses elucidate the chirality-dependent signal transduction mechanisms. A hydrogel-based chip integrated with FeCDs was fabricated and coupled with smartphone imaging and an XGBoost algorithm, enabling extraction of 1019 image features for high-accuracy quantification (R2 > 0.97, Error < 1.5%) and 100% chiral discrimination. The platform was successfully applied to complex real-world samples and enabled in vivo imaging and semiquantitative analysis of ascorbic acid distribution in an Oryzias latipes model. This work provides insight into stereoselective interactions with metal-doped carbon dots and offers an intelligent, portable tool for chiral sensing and physiological tracing.