Interpretable AI Reasoning for the Identification of Vibrational Spectroscopic Markers of Acetaminophen Impurities

This study evaluates the capability of the Gemini 3 artificial intelligence model to identify and distinguish the active pharmaceutical ingredient Paracetamol (PCA) from its main synthetic impurities, p-aminophenol (PAP) and \iupac{p-nitrophenol} (PNP), based on reasoning grounded in vibrational spectroscopic data. To this end, vibrational mode tables obtained from theoretical calculations using Density Functional Theory (DFT) at the M$06$-$2$X/$6$-$311++$G(d,p) level of theory were employed. The model analyzed diagnostic markers distributed across three distinct spectral regions ($200$ to $4000$ $cm^{-1}$), correlating specific structural variations with their corresponding vibrational signatures. The results demonstrate that Gemini 3 can associate topological and functional differences among molecules with characteristic spectroscopic patterns, yielding interpretable and consistent molecular representations. Consequently, this study highlights the potential of artificial intelligence models as auxiliary tools for automated pharmaceutical quality control, contributing to the reliable identification of impurities in synthetic drug pathways.