# Machine Learning-Based Biomarker Discovery from Serum Trace Elements and Biochemical Parameters in Patients with Nasal Polyps

## TL;DR

This study uses machine learning to find blood markers for nasal polyps, identifying zinc, copper, glucose, and inflammation indicators as potential non-invasive diagnostic tools.

## Contribution

The novel use of machine learning and multivariate analysis to identify serum trace elements and biochemical parameters as potential biomarkers for nasal polyps.

## Key findings

- Nasal polyp patients had significantly lower zinc levels and higher copper and glucose compared to healthy controls.
- Logistic regression achieved 100% classification accuracy, with zinc, copper, glucose, and Zn/Cu ratio identified as key features.
- Parameters like Cu, Zn/Cu ratio, glucose, and inflammatory indices showed strong discriminative power for NP diagnosis.

## Abstract

Nasal polyps (NP) are benign mucosal outgrowths associated with chronic inflammation that can significantly reduce quality of life. This study aimed to evaluate changes in inflammation, oxidative stress, and trace element homeostasis in NP patients and to identify potential non-invasive diagnostic biomarkers. A total of 22 patients with NP and 19 healthy individuals were included in the study. Serum levels of trace elements, including zinc (Zn), copper (Cu), and selenium (Se), were measured using inductively coupled plasma mass spectrometry (ICP-MS). Biochemical parameters including white blood cell count (WBC), red blood cell count (RBC), platelet count (PLT), eosinophils (EO), hemoglobin (HGB), glucose, creatinine, alanine aminotransferase (ALT), and thyroid-stimulating hormone (TSH) were assessed, along with inflammatory indices such as neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR). Data were analyzed using classical statistical methods, including the Shapiro–Wilk test, independent samples t-test, Mann–Whitney U test, and receiver operating characteristic (ROC) analysis. Multivariate analyses such as principal component analysis (PCA), orthogonal partial least squares discriminant analysis (OPLS-DA), and variable importance in projection (VIP) scoring were performed. In addition, machine learning algorithms including Naive Bayes, support vector machines (SVM), random forest, k-nearest neighbors (KNN), and logistic regression were employed. SHapley Additive exPlanations (SHAP) analysis was used to interpret the most influential features of the best-performing model. Compared to controls, NP patients exhibited significantly higher levels of WBC, Cu, glucose, and NLR along with significantly lower levels of Zn, PLR and the Zn/Cu ratio. Specifically, the mean Zn level was 2130.974 ± 3516.317 µg/mL in the NP group versus 11,331.127 ± 27,697.378 µg/mL in controls (p = 0.018). Cu (AUC = 0.866), glucose (AUC = 0.777), and WBC (AUC = 0.748) showed strong discriminative power. OPLS-DA revealed clear group separation, highlighting Cu, Zn/Cu, glucose, Se, and PLR as high-impact variables. Optimized logistic regression achieved 100% classification accuracy, with SHAP analysis confirming Zn, Zn/Cu, Cu, and glucose as the most influential features. These preliminary findings suggest that inflammation, trace element imbalance, and metabolic alterations can be detected biochemically in NP patients. Parameters such as serum Zn and Cu levels, Zn/Cu ratio, glucose, and inflammatory indices may serve as promising non-invasive diagnostic biomarkers. Further validation in larger and independent cohorts is warranted before clinical implementation.