Machine learning identifies shared blood transcriptional biomarkers and immune correlates across antiphospholipid syndrome and systemic sclerosis

Antiphospholipid syndrome (APS) and systemic sclerosis (SSc) are immune-mediated multisystem autoimmune diseases with distinct clinical phenotypes but overlapping pathogenic themes, including immune dysregulation, chronic inflammation, and endothelial injury. Using peripheral blood transcriptome datasets from the Gene Expression Omnibus (GSE102215: 9 APS/9 controls; GSE231691: 49 SSc/18 controls), we performed differential expression analysis within each cohort (limma; |log2FC|>1, P<0.05) and identified 281 genes dysregulated in the same direction in both diseases (100 upregulated and 181 downregulated). Enrichment analyses highlighted interferon-related and cytokine/inflammatory signaling programs in APS and SSc. To derive a compact diagnostic signature, we combined random forest feature ranking with a single-hidden-layer artificial neural network, prioritizing five shared candidate biomarkers (S100A8, IER5L-AS1, LTK, PRR5-ARHGAP8, and PCDH1). Each gene showed consistent case-control differences in both cohorts (P<0.001) and achieved good discrimination (AUC>0.75), with S100A8 performing most consistently (AUC=0.98 in APS; AUC=0.88 in SSc). CIBERSORT deconvolution indicated a myeloid-skewed blood profile in both diseases, characterized by higher neutrophil and monocyte/macrophage signals; SSc additionally showed stronger inferred CD4+ T cell and NK cell signals. S100A8 expression correlated with inferred neutrophil abundance in both cohorts (APS r=0.62; SSc r=0.58; P<0.05). Finally, miRNA-target prediction and DSigDB drug-signature enrichment generated regulatory and pharmacologic hypotheses, including immune-regulatory miRNAs (e.g., miR-155 and miR-146a) and candidate compounds (celecoxib, tamibarotene, HMN-176, and XMD14-99). Overall, these results nominate shared blood transcriptional markers and immune correlates across APS and SSc for follow-up validation.