A Systematic Multi-LLM AI Framework for Immunotherapy Biomarker Discovery and Target Identification

Immune checkpoint inhibitors have revolutionized cancer therapy, yet their efficacy is frequently limited by primary and acquired resistance. This underscores the need for novel immune regulatory targets that can improve or broaden patient responsiveness to immunotherapy. Among these, immunoglobulin (Ig) domain-containing proteins play central roles in immune signaling and tumor immune evasion, but many remain underexplored as therapeutic targets. Although recent advances in RNA sequencing, single-cell technologies, and multi-omics have expanded our understanding of gene and protein expression, integrating these data into actionable therapeutic insights remains a major challenge. Recent developments in Artificial Intelligence (AI), particularly in natural language processing and large language models (LLMs), offer a promising solution. Models such as GPT-4o, Llama 3.1-8B, and Gemini 1.5 Flash have demonstrated exceptional reasoning capabilities in biomedical applications. However, their use in immunotherapy biomarker discovery has yet to be fully realized. In this study, we introduce a novel AI-based framework that employs a multi-LLM approach to systematically identify and prioritize candidate biomarkers from a curated set of Ig-domain-containing genes. By integrating multi-omics data with structured prompt engineering and comparative model reasoning, our platform enhances the robustness, reproducibility, and interpretability of gene selection while minimizing bias inherent to single-model predictions. To our knowledge, this is the first study to apply a multi-LLM strategy to immunotherapy biomarker discovery. Our findings support the broader utility of LLMs in precision oncology and highlight their potential to accelerate the identification of novel, clinically actionable targets.