Machine learning-assisted colorimetric sensor array for identification and quantitative analysis of iron chelators

Simultaneous monitoring of multiple iron chelator levels is essential for guiding dose adjustments to prevent transfusion-induced iron overload in thalassemia patients. Although mimetic enzyme-based colorimetric assays are attractive, the multi-chelator cross-interference and the poor activity of most mimetic enzymes at near-neutral pH severely hinder their multiple iron chelator monitoring. Herein, we report a Ferric Nitrilotriacetate/Polyoxometalate (FeNTA/POM) dual-substrate colorimetric sensor array integrated with machine learning for simultaneous identification and quantification of multiple iron chelators. POM acts as a cocatalyst to promote electron transfer and Fe³⁺/Fe²⁺ cycling, ensuring excellent peroxidase-like activity of the sensor array at near-neutral pH, as reflected by the V_max values of FeNTA/POM being 7.1-fold and 4.9-fold higher than those of FeNTA for 3,3',5,5'-tetramethylbenzidine (TMB) and H₂O₂, respectively. The three clinical chelators, namely deferoxamine (DFO), deferasirox (DFX), and deferiprone (DFP), exhibit distinct iron-chelating capacities (DFO > DFX > DFP). By disrupting the Fe³⁺/Fe²⁺ redox cycle to varying degrees, they suppress the peroxidase-like activity of FeNTA/POM, thereby producing unique colorimetric responses and characteristic fingerprints in the presence of H₂O₂ and chromogenic substrates. By incorporating machine learning algorithms, this sensor array enables high-precision discrimination and quantitative prediction of multi-chelator levels in human serum samples with good recoveries ranging from 90.9 % to 108.1 %, which can effectively avoid the cross-interference in conventional single-signal readout methods. This mimetic enzyme-based sensor array is simple to operate, cost-effective, and capable of multi-component recognition, offering a new methodological framework for iron chelators and intelligent colorimetric analysis of complex drug systems.