Exploratory Machine Learning Prioritizes Shared Blood Transcriptional Candidate Genes and Immune Correlates Across Antiphospholipid Syndrome and Systemic Sclerosis

Antiphospholipid syndrome (APS) and systemic sclerosis (SSc) are distinct autoimmune diseases, but both are characterized by immune dysregulation, persistent inflammation, and vascular injury. To explore whether they also share blood transcriptional features, we analyzed peripheral blood datasets from GEO (GSE102215, APS: nine patients/nine controls; GSE231691, SSc: 49 patients/18 controls). Differential expression analysis performed separately in each cohort using edgeR with cohort-specific fold-change thresholds and nominal p < 0.05 identified 281 genes that were altered in the same direction in both diseases, including 100 upregulated and 181 downregulated genes. Enrichment analyses in both cohorts consistently pointed to interferon-related, cytokine, and inflammatory pathways. We then combined random forest and a single-hidden-layer artificial neural network as exploratory feature-prioritization approaches and identified five shared candidate genes: S100A8, IER5L-AS1, LTK, PRR5-ARHGAP8, and PCDH1. All five genes showed significant case-control expression differences in both datasets (p < 0.001). Given the small APS cohort, the machine-learning analyses were considered exploratory and were not interpreted as evidence of validated diagnostic performance. CIBERSORT suggested that both diseases have a blood immune profile enriched for myeloid signals, particularly neutrophils and monocyte/macrophage populations, whereas SSc showed relatively stronger CD4+ T-cell and NK-cell signals. S100A8 expression was positively correlated with inferred neutrophil abundance in both datasets (APS r = 0.62; SSc r = 0.58; p < 0.05). Finally, miRNA prediction and DSigDB enrichment pointed to possible upstream regulators and candidate compounds, including miR-155, miR-146a, celecoxib, tamibarotene, HMN-176, and XMD14-99. These findings represent concordant signals observed across two parallel disease cohorts rather than independent cross-disease validation and therefore require confirmation in independent cohorts.