Assistant Professor Augusta University Augusta, Georgia, United States
Introduction/Rationale: Sarcoidosis is a chronic inflammatory disease that can damage many organs. Current treatments use broad immunosuppressants, but many patients continue to show active or recurring disease. A key problem is that the immune signals that drive harmful CD4 T cell activation are still unclear. Recent studies show that biological doublets, which are physical pairs of CD4 T cells and antigen-presenting cells (APCs), carry direct signs of antigen presentation in human blood. Current single-cell tools cannot reliably separate true biological doublets from technical artifacts. This limits progress in understanding disease-driving immune signals.
Methods: We developed an AI framework that detects biological CD4 T cell–APC doublets in single-cell datasets. Our AI model integrates gene expression, surface protein markers from CITE-Seq, and TCR signals. We trained and tested our AI model using blood samples from sarcoidosis patients. The model assigns each cell a doublet probability score and an antigen-presentation score to identify true immune synapses.
Results: Our AI model identified a consistent population of biological CD4 T cell–APC doublets with matched surface markers, shared transcriptomic features, and linked TCR signals. These doublets showed strong ligand–receptor signals linked to T cell activation. Using these cells, we defined cytokine pathways enriched in sarcoidosis. We then tested the top pathways and confirmed that specific cytokine signals drive early CD4 T cell activation.
Conclusion: This work introduces the first AI framework to detect active immune synapses in sarcoidosis. By focusing on true biological doublets, this approach provides a direct view of how CD4 T cells and APCs interact during disease. The results identify measurable immune signals in blood that can support biomarker development. They also reveal specific cytokine pathways that may guide more precise and targeted therapies.
