Quantitative detection of 6-thioguanine in body fluids based on a free-standing liquid membrane SERS substrate.

The adverse reactions caused by 6-thioguanine (6-TG) in anti-cancer treatment are closely related to the dose, leading to the urgent need for clinical monitoring of its concentration. In this work, a highly reproducible free-standing liquid membrane (FLM) surface-enhanced Raman spectroscopy (SERS) substrate was developed to detect 6-TG in human urine and serum quantitatively. Briefly, a prepared sample was adjusted to pH 2 and mixed with concentrated core-shell bimetallic nanoparticle (AgcoreAushell NP) suspension. The Au/Ag ratio of the AgcoreAushell NPs was optimized. Then the mixture was formed into an FLM using a custom mold. The relative standard deviation (RSD) of the experimental results can be stabilized below 10% (n ≥ 10). The R2 of the calibration curve in the range of 10 ~ 100 µg kg-1 was 0.988. In addition, the limit of detection (LOD) (3σ/k) of 6-TG was 5 µg kg-1. The FLM SERS platform has been successfully applied to the rapid and reliable analysis of 6-TG spiked in body fluids.

Rapid diagnosis and recurrence prediction of choledocholithiasis disease using raw bile with machine learning assisted SERS.

Surface-enhanced Raman spectroscopy (SERS) analysis based on body fluids has been widely applied in disease diagnose. Choledocholithiasis is a widespread and often recurrent digestive system disease, with limited data on factors predicting its formation and reappearance. Bile contains many components that could provide valuable diagnostic information; however, the current diagnosis of biliary disease by SERS focuses on detecting specific component in the bile, overlooking the complex interplay and correlations among multiple factors that could be crucial for accurate diagnosis. This work directly obtained multi-component SERS spectral information of raw bile from 46 patients. Characteristic information was extracted from bile SERS spectra using Principal Component Analysis (PCA), revealing variations in the content of characteristic components associated with different choledocholithiasis types and their recurrence frequency. Pearson correlation analysis was also introduced to reveal the interactions of primary active substances pertinent to choledocholithiasis diagnosis. The efficacy of PCA and Support Vector Machine (SVM) models in classifying stone types, presented an accuracy of 99.2 %. Furthermore, the interaction patterns among SERS characteristic components in choledocholithiasis recurrence frequency were revealed, and with the support of SVM, the prediction for different recurrence rates reached an accuracy of 95.2 %. Overall, this work demonstrates that integrating SERS with machine learning can support disease diagnosis and the interpretation of correlations among multiple components, facilitating elucidating the disease mechanisms.