Exploring emerging exposure scenarios via intelligent non-target screening: Chemical space characterization of pet hair contaminants

Pet ownership in urban households has created unique indoor exposure scenarios where humans and their companion animals cohabit intensively. Pet hair has become an emerging source of contaminants that may affect human health, yet the chemical composition and accumulation patterns in pet hair remain poorly characterized. Herein, an integrated data-to-laboratory framework combining text mining, machine learning, and high-resolution mass spectrometry-based non-target screening was developed to investigate the chemical space in pet hair and indoor dust systematically. A total of 16,692 publications were mined to curate 3,684 indoor-relevant chemicals from which we established a molecular ion library containing 2,661 compounds. Complementarily, a diagnostic ion library with 30 fragment ions was generated using an artificial neural network coupled with a convolutional capsule neural network. Together, the two newly established libraries were employed for non-target screening of 14 pet hair and 10 dust samples. These scenario-specific libraries substantially enhanced the coverage of chemical identification, increasing the number of annotated compounds in major chemical categories up to 28-fold and reducing analysis time by 47-fold compared with conventional target analysis. Non-target screening revealed high similarity between matrices, with over 50% of shared molecular features and consistent distributions of functional and structural categories. Artificial neural network analysis identified key physicochemical descriptors governing contaminant accumulation, including log P, pK_a, aromatic atom count, molecular flexibility, and polarizability. Further analysis of non-targeted compounds confirmed that these parameters drive the observed similarity between pet hair and dust.