Machine learning-enhanced fluorescence analysis of glucose–tryptophan interactions toward diabetes diagnostics

Noninvasive optical approaches for metabolic assessment have attracted increasing attention due to their potential for rapid, painless, and accessible disease monitoring; however, accurately quantifying and correlating multiple biomarkers remains challenging. In this study, we systematically investigate the interaction between glucose and tryptophan, two key metabolic indicators, under controlled aqueous conditions within physiologically relevant urinary concentration ranges. Tryptophan solutions were synthetically prepared in the range of 1–10 ppm, corresponding to typical urinary concentrations (∼3–8 μmol/dL), while glucose concentrations were varied between 20 and 200 ppm to reflect physiological levels (0–0.8 mmol/L). Steady-state fluorescence spectroscopy using a Xenon arc lamp revealed characteristic emission peaks at 357 nm for tryptophan and 450 nm for glucose. A statistically robust inverse relationship between glucose and tryptophan fluorescence intensities was observed, yielding a strong negative monotonic correlation (Spearman’s ρ = −0.999, p < 0.001), further validated by a strong quadratic fit (R2 ≈ 0.98). To quantitatively validate this nonlinear biochemical interplay, machine learning–based Support Vector Regression (SVR) was employed, with the radial basis function kernel achieving sub-1% prediction errors. Residual analysis and heat-map visualization confirmed the robustness and predictive reliability of the proposed framework. This work provides mechanistic validation of the glucose–tryptophan fluorescence interaction previously observed in human urine and establish a controlled proof-of-concept framework to guide future biological and clinical validation studies.