Environment Stable Prioritization of Disease Associated Myeloid Signatures in Single Cell IBD Atlases

Single cell disease signatures can appear accurate while relying on donor, tissue, platform, or study specific shortcuts that do not transfer across cohorts. This study presents an environment stable prioritization framework for identifying transferable myeloid disease associated signatures in public inflammatory bowel disease (IBD) single cell atlases. Public CZ CELLxGENE Discover resources were used to assemble a multi study intestinal myeloid training atlas and an independent colonic myeloid cohort for external evaluation. Cells were aggregated to donor level pseudobulks, disease labels were harmonized to a control versus IBD outcome, and all feature ranking was performed within the relevant training split. A baseline panel ranked genes by pooled disease control separation, whereas the environment stable panel required consistent disease control direction across training environments and applied a variance penalized stability score; a sparse variant added elastic net regularization after stable ranking. In this case study, the stability based panels transferred more effectively than the pooled baseline under both leave one environment out testing and independent external evaluation, although the external cohort remained modest in size and therefore supports only preliminary transfer evidence. Disease pseudobulks showed higher monocyte and lower macrophage proportions than controls, yet the prioritized stable genes retained their direction after composition adjustment for major myeloid subtype proportions and environment. The stable panel organized into chemokine recruitment, NF \((\kappa)\)B and inflammatory regulation, metal ion or oxidative stress, and resident like remodeling axes, supporting interpretation as transferable disease associated myeloid programs rather than as confirmed causal regulators. The framework does not identify causal drivers from observational data alone; instead, it prioritizes candidate signatures for composition adjusted sensitivity analysis, perturbation matching, and experimental validation.