[Establishment of a Noninvasive Diagnostic Model for Wilson Disease Using Metallomics and Machine Learning]

To analyze the differences in urine metal profiles between patients with Wilson disease (WD) and healthy controls, to identify early diagnostic biomarkers, and to develop a non-invasive diagnostic model using machine learning.

63 WD patients and 63 matched healthy controls were included. Urine samples and clinical data were collected from all the participants. The concentrations of 51 urine metals were determined using inductively coupled plasma mass spectrometry (ICP-MS). Differences between the two groups were compared using the Wilcoxon signed-rank test. Differential metal features were selected based on detection rates > 50%, P < 0.05 and |log₂FC| > 1, and feature selection was performed using elastic net regression. Non-metric multidimensional scaling was used to analyze the metal distribution between the two groups. Spearman correlation analysis and quantile g calculation models were used to analyze the correlation between metals and clinical indicators. A diagnostic model was developed using the random forest algorithm, and the performance was evaluated using receiver operating characteristic curves and confusion matrices.

Urine metallomics analysis revealed statistically significant differences in the levels of Cu, Zn, Ca, Co, Sr, Ti, Y, Cs, Rb, Cd and Sn between the case and control groups. Cu/Zn, Cu/Se and Zn/Se ratios were significantly higher in the case group. Elastic net regression identified 14 key features, with Cu having the largest standardized regression coefficient (β = -14.628). Non-metric multidimensional scaling confirmed the separation of metal profiles between the two groups. Correlation analysis showed significant associations between Cu, Cu/Zn, and liver function indicators, with Spearman correlation coefficients ranging from -0.58 to 0.67. The quantile g calculation model suggested that metal mixtures had a significant negative effect on the A/G ratio. The random forest model exhibited excellent diagnostic performance, with an area under the curve (AUC) of 0.99 (95% CI: 0.97-1.00) in the training set and 0.97 (95% CI: 0.94-1.00) in the testing set, outperforming single indicators.

Urine metallomics analysis indicated that the Cu/Zn ratio obtained superior diagnostic efficiency compared to traditional urine copper test. Additionally, the diagnostic model based on differential metal characteristics demonstrated high accuracy, providing a new method for the early non-invasive diagnosis of WD.