An AI-enabled multi colorimetric sensor array: Towards rapid and noninvasive detection of neuroblastoma urinary markers

In this research, a single-component multi-colorimetric sensor array was developed to simultaneously detect homovanillic acid (HVA) and vanillylmandelic acid (VMA) tumor markers based on their oxidation ability in exposure to silver ions. To this aim, gold nanorods (AuNRs) were used as optical signal transducers. The aspect ratio alterations of AuNRs in response to the concentration of the analytes were achieved by silver shell metallization on the surface of AuNRs. Under optimal conditions, absorbance spectra of the probe and associated vivid color variations were recorded, followed by analysis using a hybrid machine learning classification algorithm to discriminate between HVA, VMA, HVA: VMA along with presumptive interfering compounds (INF). Data was prepared for classification using linear discriminant analysis (LDA), by creating a new space through principal component analysis (PCA). The ideal number of PCs was determined via Leave-Many-Out cross-validation to maximize the classification performance of the model. Besides, partial least squares regression (PLSR), which maximizes covariance between different concentrations of biomarkers, was utilized for quantification. Results indicated that the proposed ML-based sensor array has the potential to diagnose childhood metabolic errors. This is supported by its capability of detecting HVA, VMA, and HVA:VMA concurrently with 100% sensitivity and specificity in human urine samples. Also, a high correlation between predicted and measured values verified the model's exceptional prediction performance with low detection limits of 0.22 μM and 0.29 μM for HVA and VMA, respectively. [Display omitted] •A single-component sensor array has been developed to detect tumor markers at the same time.•Colorimetric fingerprints were generated by AuNRs for identifying HVA and VMA.•A hybrid machine-learning approach was used for qualitative and quantitative evaluations.•The outcomes of prediction in urine sample show good potential for diagnosing metabolic issues of children.