Machine learning-assisted modular triple-signal ratiometric ECL biosensor for multi-Alzheimer's biomarkers detection

Triple-signal ratiometric electrochemiluminescence (ECL) biosensors exhibit superior multidimensional error correction capabilities compared to dual-signal systems. However, current design strategies are constrained by the lack of established criteria for ratiometric signal selection and limited adaptability to multiple targets and detection modes, which highlights the urgent demand for more versatile design strategies.

Herein, a modular triple-signal ratiometric ECL biosensor was fabricated. Integrated with machine learning (ML) algorithms, it enabled accurate identification and quantification of different Tau protein, successfully distinguishing Alzheimer's disease (AD) patients from healthy controls and comorbidity populations. The biosensor incorporated graphitic carbon nitride (g-C₃N₄), luminol, and Ru(bpy)₃²⁺ as luminophores, generating multiple potential-resolved ECL signals at both cathodic and anodic potentials. Specifically, both g-C₃N₄ and luminol utilized H₂O₂ as the co-reactant, exhibiting ECL emission at -1.5 V and +0.5 V, respectively, whereas Ru(bpy)₃²⁺ generated emission at +1.2 V with TPrA as the co-reactant. The distinct coreaction pathways under different potentials ensured that the three emission channels were mutually independent and free from cross-interference. Through a machine learning algorithm based on variable importance in projection scores, nine ratiometric combinations were ranked, and the optimal ratio was independently validated by receiver operating characteristic analysis. This modular biosensor was configured in both signal-on and signal-off modes, and was designed to detect Tau 381 and Tau 441 at the optimal ratio with detection limits of 3.8 and 3.7 fg mL^-1, respectively.

This modular biosensor overcame case-by-case optimization and provided a versatile platform for multi-target sensing with significant clinical application potential.