BIOGEN: Evidence-Grounded Multi-Agent Reasoning Framework for Transcriptomic Interpretation in Antimicrobial Resistance

arXiv:2510.16082v4 Announce Type: replace-cross Abstract: Interpreting gene clusters derived from RNA sequencing (RNA-seq) remains a persistent challenge in functional genomics, particularly in antimicrobial resistance studies where mechanistic context is essential for downstream hypothesis generation. Conventional pathway enrichment methods summarize co-expressed modules using predefined functional categories, but they often provide limited coverage and do not yield cluster-specific mechanistic explanations grounded in primary literature. We present BIOGEN, an evidence-grounded multi-agent framework for post hoc interpretation of RNA-seq transcriptional modules that integrates biomedical retrieval, structured interpretation, and multi-critic verification. BIOGEN organizes knowledge from PubMed and UniProt into traceable cluster-level explanations with explicit evidence reporting and confidence tiering. On the primary Salmonella enterica dataset, BIOGEN achieved strong evidence grounding and biological coherence, with a BERTScore of 0.689, RAGAS Faithfulness of 0.930, Semantic Alignment Score of 0.715, and KEGG Functional Similarity of 0.342. All retrieval-grounded configurations maintained a hallucination rate of 0.000, compared with 0.100 for the LLM-only baseline. Across four additional bacterial RNA-seq datasets, BIOGEN preserved zero hallucinations and provided broader thematic coverage than KEGG/ORA-based enrichment. Comparative experiments with representative agentic AI baselines further show that retrieval access alone is insufficient to ensure traceable biological interpretation, highlighting the importance of coordinated evidence grounding and verification in biomedical reasoning.