JAK-Centric Explainable Few-Shot Gene-Expression Diagnosis Framework for Alopecia via MultiPLIER Priors and Relation-Style Set-to-Set Comparison

The study presents a novel diagnostic framework for alopecia areata (AA) using a few-shot deep learning classifier that integrates bulk and single-cell RNA-seq data to analyze JAK–STAT signaling. Researchers identified 257 differentially expressed genes in AA, with four key genes—GZMA, IL2RB, IL2RG, and EOMES—validated as biomarkers. The framework uses MultiPLIER latent variables and a Relation-style set-to-set comparator to distinguish AA from controls, even with limited samples, while maintaining interpretability. Functional validation confirmed the activation of the IL2RB/IL2RG–EOMES–GZMA axis in AA, linked to JAK inhibitors and cyclosporine, unlike androgenetic alopecia (AGA), which lacks JAK–STAT perturbation. This approach offers a non-invasive biomarker and a target for precision JAK blockade, demonstrating an XAI-driven method for small-cohort genomic diagnosis.