Molecular-Level Recognition Surfaces Drive Machine-Learning Discrimination of Homologous Gases

Chemoresistive semiconductor sensing empowered by machine learning offers a promising route for accurate discrimination and quantification of chemically homologous gases, but remains challenging due to overlapping adsorption behaviors during recognition and presenting nearly identical electronic responses. Here, we design molecular-level recognition surfaces to modulate reaction sites and amplify the subtle differences in homologous gas responses, which is critical for machine-learning-assisted discrimination. By conformally integrating near-bi-molecular-layer amorphous phosphotungstic acid (∼2 nm thick) onto WO₃ nanowires, we generate a continuum of Lewis and Brønsted acid sites with diverse adsorption activation energies and differential reaction pathways toward ammonia/amine homologous gases. This design produces characteristic fingerprint spectra based on temperature-dependent steady-state response intensity and transient reaction rate, achieving a high recognition accuracy of 91.0% for four target ammonia/amines using a minimal training dataset and simple classifiers. Remarkably, this sensing system quantitatively resolves binary ethylamine/dimethylamine isomer mixtures with high precision (R² > 0.99, error < 4%) with excellent generalization, a capability not realized previously. The near molecular-level amorphization strategy bridges the gap between single-atom dispersion and conventional nanoheterostructures, providing a powerful platform to modulate gas recognition processes and driving machine-learning discrimination and quantitative analysis of homologous gases in complex environments.