IMMUNIA: A Reasoning-Centered Framework for Immunoregulatory Surfaceome Discovery

High Resolution Image Download MS PowerPoint Slide Biomarker discovery in immunotherapy remains limited by approaches that rely primarily on correlation-based analyses, which often fail to capture the context-dependent and mechanistic nature of tumor immune interactions. Here, we present IMMUNIA, a reasoning-centered framework designed to prioritize immunoregulatory surfaceome genes through structured, interpretable, and multimodel inference. IMMUNIA integrates standardized prompting, literature-grounded context, and consensus reasoning across multiple large language models to evaluate candidate genes along key immunological dimensions, including immunotherapy relevance, inflammation, and NF-κB signaling. Applied to transcriptomic data from prostate cancer, IMMUNIA systematically analyzed 458 immunoglobulin domain-containing surfaceome genes and identified a prioritized set of candidates through multirun, cross-model evaluation. Internal validation using positive and negative control genes demonstrated robust discrimination of immune-relevant targets, while cross-model agreement and low variability across repeated runs supported the stability of the framework. Consensus prioritization recovered established immunoregulatory molecules, including IL1R1, CD276, and B2M, and further highlighted PTPRS, VCAN, and MXRA5 as candidate genes with potential roles in stromal-mediated immune regulation. The biological interpretations presented in this study are grounded in prior literature and reflect expert-level mechanistic reasoning, with IMMUNIA serving to systematically structure and scale this reasoning process. Rather than generating de novo biological claims, this framework enables the integration of existing knowledge into testable hypotheses, providing a transparent and reproducible path from transcriptomic data to mechanistically informed biomarker prioritization. These findings suggest that reasoning-centered artificial intelligence can complement conventional data-driven approaches and support the discovery of candidate immunoregulatory targets within the tumor microenvironment.