Photoresponsive HOF-Based Platforms for Large Language Model-Assisted Multimodal Diagnosis of Metabolic Diseases

Artificial intelligence (AI), particularly large language models (LLMs) such as chat generative pretrained transformer (ChatGPT), is revolutionizing various fields. Here, we present an AI-enhanced olfactory diagnosis platform that integrates a photoresponsive hydrogen-bonded organic framework (HOF)-based sensor with the multimodal GPT-4o model. Eu-FDA@HOF prepared by introducing europium ions (Eu³⁺) and 2,5-furandicarboxylic acid (FDA) via a host-guest coassembly strategy possesses ultralong room-temperature phosphorescence (461 ms) and characteristic fluorescence emission. The photoresponsive sensor achieves selective and rapid detection of phenylpyruvic acid (PPA) and creatinine (Cr), odor molecules of phenylketonuria and renal dysfunction, respectively, with low detection limits (1.53 μM and 1.18 μM). Body odor, as a reflection of metabolic status, offers a potential foundation for olfactory diagnosis. Leveraging PL responses, GPT-4o, guided via a human-in-the-loop strategy, performs accurate concentration prediction and generates clinically interpretable diagnoses. This work bridges photoluminescent sensing and AI-driven medical interpretation, demonstrating a versatile and accessible platform for intelligent olfactory diagnostics.