A machine learning-driven dip-searching algorithm for angle-resolved Surface Plasmon Resonance (aSPR) biosensing
Journal
Sensors and Actuators A: Physical
Journal Volume
395
Start Page
117104
ISSN
09244247
Date Issued
2025-12-01
Author(s)
Abstract
Angle-resolved Surface Plasmon Resonance Biosensor is widely used for biomolecular detection due to its high sensitivity and linear response to changes in refractive index. However, conventional angle curve fitting algorithms exhibit significant variations due to situations like low signal-to-noise ratios, inherent incident angle differences in the high-throughput format, variations in surface chemical states, and limitations in the mechanical scanning resolution of angular sensorgrams. To address these challenges, we propose a novel adaptive dip-searching algorithm that employs a four-layer decision tree regression structure with feature extraction, asymmetry truncation, domain selection, and fine-tuning functions, which can segment the best linear fitting range near the dip angle in the derivative format of the angle spectrum to determine the optimum dip position without user intervention. In comparison to conventional polynomial fitting (4.24 mdeg) and centroid methods (180.81 mdeg), our method achieves a significantly lower dip estimation error (2.8 mdeg). Performance evaluation was conducted using simulated data generated with a Fresnel multilayer reflectivity model that accounted for variations in film thickness (40–55 nm), incident angles (±0.35°, 54.15°–54.85°), and SNR levels (40–60 dB). Importantly, the algorithm supports sub-millisecond processing per spectrum, making it suitable for real-time applications. Experimental validation using protein–protein interaction data further confirmed its effectiveness, achieving 80.65 % accuracy in angle shift determination—approximately 1.8 times higher than the 45.16 % accuracy achieved by polynomial fitting.
Subjects
Angle-resolved SPR
Decision Tree Regression (DTR)
Dip angle recognition
MultiROIs
Surface Plasmon Resonance (SPR)
Publisher
Elsevier B.V.
Type
journal article