A Symbiotic AI Approach for Trustable and Interactive Microbiome Data Interpretation

The transition from raw genomic data to actionable clinical insights remains a primary challenge in microbiome-based diagnostics. While modern information systems can effectively identify taxonomic anomalies and provide quantitative visualizations (e.g., phyla abundance ratios), the resulting reports often present a “semantic gap” for patients and General Practitioners, who may lack specialized expertise. This position paper presents a Symbiotic AI (SAI) approach to bridge this gap by integrating Large Language Models (LLMs) with existing visualizations within a trustworthy and reproducible pipeline. Aligning with the AI2VIS4BigData Reference Model, our approach moves beyond static reporting by utilizing an LLM-based Interpretative Layer. To ensure reproducibility, we propose a semantic intermediate mapping that structures clinical data into a machine-readable format before the agent processes it. To ensure trust, the system employs an internal grounding process that synthesizes narrative explanations from validated clinical snippets and laboratory thresholds. Moreover, an interactive symbiosis is facilitated through a conversational interface linked to existing genomic visualizations. This interface enables a bi-directional exchange: the AI provides context-aware summaries of dysbiosis patterns, while the practitioner can conduct further inquiries and provide corrective feedback via the UI. The system captures these human-in-the-loop interactions in a vector-based expert knowledge base, enabling it to refine future explanations based on real-world clinical experience without compromising model stability. The proposed approach emphasizes traceability and interpretative clarity and may serve as an exemplar for regulatory-aware intelligent visualization systems.