Mei Wang,1,* Jing Wang,2,* Fei Lv,3 Aifeng Song,1 Wurihan Bao,1 Huiyun Li,1 Yongsheng Xu3 

1Department of Rheumatology and Immunology, Inner Mongolia Autonomous Region People’s Hospital, Hohhot, Inner Mongolia, 010017, People’s Republic of China; 2Department of Rheumatology and Immunology, The Affiliated Hospital of Inner Mongolia Medical University, Hohhot, Inner Mongolia, 010050, People’s Republic of China; 3Orthopedic Center, Inner Mongolia Autonomous Region People’s Hospital, Hohhot, Inner Mongolia, 010017, People’s Republic of China

*These authors contributed equally to this work

Correspondence: Huiyun Li, Department of Rheumatology and Immunology, Inner Mongolia Autonomous Region People’s Hospital, Hohhot, Inner Mongolia, 010017, People’s Republic of China, Email Lihuiyun2572@163.com Yongsheng Xu, Orthopedic Center, Inner Mongolia Autonomous Region People’s Hospital, Hohhot, Inner Mongolia, 010017, People’s Republic of China, Email myfs202406@163.com

Background: Growing research has demonstrated that alterations in Th2 and Th17 cell composition were linked to systemic juvenile rheumatoid arthritis (sJRA). Nevertheless, whether these associations indicate a causal link remains unclear, and the potential effects of Th2/Th17-related molecules have not been clarified.
Methods: Mendelian randomization (MR) alongside transcriptome examination was implemented to ascertain the links between the Th2/Th17 cells and sJRA. Subsequently, we established an innovative machine learning (ML) framework encompassing 12 ML approaches and their 111 permutations to generate a unified Th2/Th17 classifier, which underwent verification across three separate cohorts. The hub Th2/Th17-related genes’ level in the sJRA patients was substantiated via qRT-PCR. Lastly, the SHapley Additive exPlanations (SHAP) in conjunction with the XGBoost algorithm to pinpoint ideal Th2/Th17-linked genes.
Results: Based on MR analyses of two sJRA GWAS, 2 immunophenotypes (lymphocyte and IgD+ B cell) were causally linked to sJRA. Based on IOBR algorithms, we revealed that lymphocyte Th2/Th17 proportion was markedly changed in sJRA from seven cohorts. WGCNA and differential analysis in two merged GEO cohorts identified 64 Th2/Th17-related genes. Based on the average AUC (0.844) and model stability in four cohorts, we converted 12 ML techniques into 111 combinations, from which we chose the optimal algorithm to generate an ML-derived diagnostic signature (Th2/Th17 classifier). qRT-PCR verified results. Moreover, immune cell infiltration and functional enrichment analysis suggested hub Th2/Th17-related gene potentially mediated sJRA onset. XGBoost algorithm and SHAP detected HRH2 as crucial genetic markers, which may be an important target for sJRA.
Conclusion: A diagnostic model (Th2/Th17 classifier) via 111 ML algorithm combinations in six independent cohorts was generated and validated, which stands as an effective instrument for sJRA detection. The identification of essential immune components and molecular cascades, along with HRH2, could emerge as vital therapeutic targets for sJRA intervention, providing an enhanced understanding of its fundamental processes.

Keywords: systemic juvenile rheumatoid arthritis, Th2/Th17 cells, machine learning approaches, Mendelian randomization, transcriptome