Establishment and validation of an immune-related genes diagnostic model and experimental validation of diagnostic biomarkers for autoimmune thyroiditis based on RNA-seq.

Background: Autoimmune thyroiditis (AIT), a common autoimmune disease, is a complex disease with an increasing incidence and an unknown pathogenesis that awaits the refinement of diagnostic methods and identification of diagnostic biomarkers to improve screening to identify patients at high risk of AIT earlier and provide the potential effective therapeutic drugs.

Patients And Methods: All samples for this study were from a cross-sectional survey, which was conducted among adults in two regions of Anhui Province, China. Ten representative samples (n = 5, n = 5) were selected for RNA sequencing to build a training-set in order to identify immune-related differentially expressed genes (IRDEGs), and least absolute shrinkage and selection operator (LASSO) regression analysis was subsequently adopted to screen key IRDEGs and construct a diagnostic model. Then, a test-set was created by downloading AIT transcriptome datasets from the Gene Expression Omnibus (GEO) database. The diagnostic model was systematically evaluated in the training-set and test-set by principal component analysis (PCA) analysis, receiver operating characteristic (ROC) curve and immune infiltration analysis. To identify diagnostic biomarkers, quantitative reverse transcription PCR (RT-qPCR) was used to measure the expression levels of the diagnostic genes in 80 samples. The diagnostic and therapeutic values of the diagnostic genes for AIT were investigated using gene set variation analysis (GSVA), ROC curve, logistic regression analysis and drug docking.

Results: The diagnostic model included three diagnostic genes (FGFR2, CCR1, IL1B). All ROC curves (AUC > 0.7) results suggested that the diagnostic model and the diagnostic genes had reliable predictive power. The results of logistic regression analysis showed that the three diagnostic genes were significant for AIT. The results of GSVA and immunoinfiltration analysis demonstrated that the diagnostic genes have significant negative or positive regulatory effect in immune mechanisms of AIT and the diagnostic model implements immune-related prediction algorithms. Finally, the small molecular compounds (Acetaminophen and Albuterol) were screened as the potential therapeutic drugs for AIT.

Conclusion: Using machine learning and bioinformatics techniques, this study developed and validated an AIT diagnostic model, explored the diagnostic model's prediction mechanism, verified three potential diagnostic biomarkers by experiments and predicted two small molecule therapeutic drugs.