HueBase AI: an artificial intelligence-integrated cheminformatics pipeline for automated colorimetric sensor discovery

Colorimetric sensor arrays offer a rapid, low-cost route to analyte detection across pharmaceuticals, agriculture, environmental monitoring, and food safety, classifying complex chemical mixtures through collective optical responses rather than specific molecular recognition. Their design, however, remains an empirical and time-intensive process that relies on iterative trial-and-error screening of dye-analyte combinations. Here we present HueBase AI, an end-to-end artificial-intelligence pipeline that unifies automated literature mining, cheminformatics, and machine learning to streamline colorimetric sensor discovery. A large-language-model agent, integrated with optical character recognition (OCR) and computer-vision modules, curated the first comprehensive open-source colorimetric sensing database (HueBase), comprising more than 14,000 experimentally validated dye-analyte-solvent interactions extracted from over 200 peer-reviewed publications. Each chemical entity is encoded as Morgan (2048-bit) and Molecular ACCess System (MACCS) (167-bit) molecular fingerprints, together with their concatenated representation, enabling systematic benchmarking of 16 classification algorithms across all three featurization (MACCS-only, Morgan-only, and concatenated MACCS + Morgan). An optimized XGBoost model, selected through a three-stage hyperparameter search, predicts dye-analyte responsiveness with 87.8% accuracy and was experimentally validated against two plant volatile organic compound (VOC) biomarkers, (E)-2-hexenal and methyl salicylate, using both existing reference dyes and novel candidate dyes identified from the Max A. Weaver Dye Library containing ~98,000 dyes. By transforming scattered, heterogeneous literature into a machine-actionable knowledge base, HueBase AI establishes a data-driven foundation for intelligent colorimetric sensing and charts a path toward rapid in silico dye screening and the automated design of next-generation sensor arrays.