DNA-Mediated Self-Assembled SERS Platform Combined with Machine Learning for Therapeutic Drug Monitoring

Therapeutic drug monitoring (TDM) critically requires rapid, sensitive, and practical methods for personalized medicine. Herein, we present a label-free surface-enhanced Raman spectroscopy (SERS) platform based on self-assembly of T12 DNA on Ag@ITCNPs mediated by Ca2+. This substrate generates abundant electromagnetic hotspots via electrostatic crosslinking and utilizes the intrinsic thymine peak as an internal standard to effectively eliminate serum matrix interference, achieving an analytical enhancement factor (AEF) of 5.93 × 1013. The platform enables ultrasensitive detection of various antiepileptics and immunosuppressants (e.g., cyclosporine A, carbamazepine, valproate) at the fg/mL level, with excellent reproducibility (RSD = 4.10%). By integrating unsupervised algorithms (PCA, t-SNE) and supervised models (CNN, SVM, KNN, LDA, RF, RNN, XGBoost), it achieves near 100% accuracy in discriminating between different drugs and even among individual patients receiving the same medication. Furthermore, regression models based on SERS enable precise quantification, with results showing excellent agreement with the LC−MS/MS gold standard. This work demonstrates that the synergy between self-assembly directed by DNA and machine learning provides a powerful strategy for constructing highly sensitive and selective SERS platforms, offering a promising solution for rapid therapeutic drug monitoring at the point of care in complex biological matrices.