Machine Learning and DFT Decoding Lanthanide-Doped Lignin Carbon Dots: Multicolor Sensor Arrays for Ultrasensitive Profiling Biothiols

Abstract

Biothiols are key small molecules that maintain the redox balance in living organisms, and their abnormal levels are closely associated with various diseases. The development of rapid and accurate detection methods is of great significance to precision medicine. However, traditional detection methods often rely on a single signal output, making it difficult to effectively distinguish between structurally similar biothiol congeners, and they are susceptible to interference from complex biological matrices. In this study, four types of lanthanide (Nd, Pr, La, Sm)-doped fluorescent carbon dots were synthesized via a one-step hydrothermal method using lignin extracted from corn straw as a green carbon source. The abundant functional groups on the surface of these carbon dots generate differential interactions (coordination, hydrogen bonding, electrostatic interactions) with various thiol molecules, forming characteristic fluorescence response fingerprints. Furthermore, machine learning algorithms were employed to model and classify the multi-dimensional response data collected from the four-channel sensor array. The results demonstrate that the sensor array exhibits a detection limit as low as 0.42 nM for five biothiols (GSH, Cys, Hcy, NAC, DTT), with LODs ranging from 0.42 to 3.63 nM across the four carbon dots. The random forest model achieved 98.2% accurate identification on the test set, and the spiked recovery rates in serum samples ranged from 92% to 106%. Meanwhile, density functional theory (DFT) calculations preliminarily revealed that the fluorescence quenching mechanism is predominantly governed by photoinduced electron transfer (PET), validating the modulating effect of lanthanide ions on the energy levels of the carbon dots at the atomic level. This work not only provides a new pathway for the high-value utilization of agricultural waste but also opens up a versatile sensing platform for the rapid and intelligent identification of similar analytes in complex samples.