Machine learning-based prediction of diagnostic markers for Graves' orbitopathy.

PURPOSE

The pathogenesis of Graves' orbitopathy/thyroid-associated orbitopathy (TAO) is still unclear, and abnormal DNA methylation in TAO has been reported. Thus, selecting and exploring TAO biomarkers associated with DNA methylation may provide a reference for new therapeutic targets.

METHODS

The TAO-associated expression data and methylation data were downloaded from The Gene Expression Omnibus database. Firstly, weighted gene co-expression network analysis was used to obtain the TAO-related genes, which were intersected with differentially methylated genes (DMGs), and differentially expressed genes between TAO samples and normal samples to obtain TAO-associated DMGs (TA-DMGs). Thereafter, the functions of the TA-DMGs were analyzed, and diagnostic markers were screened by least absolute shrinkage and selection operator (Lasso) regression analysis and support vector machine (SVM) analysis. The expression levels and diagnostic values of the diagnostic markers were also analyzed. Furthermore, single gene pathway enrichment analysis was performed for each diagnostic marker separately using gene set enrichment analysis (GSEA) software. Next, we also performed immune infiltration analysis for each sample in the GSE58331 dataset using the single-sample GSEA algorithm, and the correlation between diagnostic markers and differential immune cells was explored. Lastly, the expressions of diagnostic markers were explored by quantitative real-time polymerase chain reaction (qRT-PCR).

RESULTS

A total of 125 TA-DMGs were obtained. The enrichment analysis results indicated that these TA-DMGs were mainly involved in immune-related pathways, such as Th1 and Th2 cell differentiation and the regulation of innate immune response. Moreover, two diagnostic markers, including S100A11 and NKD2, were obtained by Lasso regression analysis and SVM analysis. Single gene pathway enrichment analysis showed that S100A11 was involved in protein polyufmylation, pancreatic-mediated proteolysis, and NKD2 was involved in innate immune response in mucosa, Wnt signaling pathway, etc. Meanwhile, immune cell infiltration analysis screened 12 immune cells, including CD56 dim natural killer cells and Neutrophil cells that significantly differed between TAO and normal samples, with the strongest positive correlation between NKD2 and CD56 dim natural killer cells. Finally, the qRT-PCR illustrated the expressions of NKD2 and S100A11 between normal and TAO.

CONCLUSION

NKD2 and S100A11 were screened as biomarkers of TAO and might be regulated by DNA methylation in TAO, providing a new reference for the diagnosis and treatment of TAO patients.