Machine learning-assisted photoelectrochemical aptasensor for nalbuphine monitoring using a BiOBr/SrWO4 S-scheme heterojunction.

Precise therapeutic drug monitoring is essential for optimizing nalbuphine treatment in uremic pruritus, yet rapid and sensitive analytical tools suitable for clinical application remain limited. Herein, we report a machine learning-guided photoelectrochemical (PEC) aptasensor for nalbuphine detection based on a BiOBr/SrWO4 S-scheme heterojunction. Clinical data from 102 peritoneal dialysis patients were first analyzed using seven machine learning algorithms to identify determinants of therapeutic response. Gradient Boosting exhibited the best predictive performance, with an accuracy of 84.6%, a response recall of 83.3%, and an AUC of 0.868 on the independent test set, and nalbuphine dose emerged as the most influential variable, establishing the clinical need for accurate drug monitoring. Guided by this finding, a signal-on PEC aptasensor was constructed by integrating a BiOBr/SrWO4 heterojunction photoactive interface with an NH2-modified ssDNA aptamer for selective nalbuphine recognition. The optimized BiOBr/SrWO4-40% composite displayed enhanced visible-light harvesting, suppressed charge recombination, and accelerated interfacial charge transfer. Ultraviolet photoelectron spectroscopy and density functional theory calculations further verified the S-scheme charge-transfer pathway and the built-in electric field-driven carrier separation mechanism. Under optimal conditions, the aptasensor exhibited a linear response toward nalbuphine from 0.01 to 1.2 μg mL-1 with a limit of detection (LOD) of 3.5 ng mL-1, along with good selectivity, stability, and reproducibility. In spiked serum samples, recoveries ranged from 96.05% to 99.59% with relative standard deviations below 5%. This work provides a clinically oriented strategy for integrating machine learning with PEC biosensing and offers a promising platform for personalized therapeutic drug monitoring.