High-precision identification of dihydrouracil, glycine anhydride, and piperazine using terahertz spectroscopy

Distinguishing structurally similar pharmaceutical compounds remains a significant challenge in drug quality assurance, especially when these compounds share overlapping physicochemical properties. This study focuses on three such compounds: dihydrouracil (DHU), glycine anhydride (GA), and piperazine (PIP). DHU and GA are structural isomers, while PIP features a distinct heterocyclic structure, thus providing a rigorous test of the method’s specificity. Traditional analytical techniques, such as high-performance liquid chromatography (HPLC) and nuclear magnetic resonance (NMR), are often limited in rapid on-site deployment due to the need for sample pretreatment and lengthy analysis times. Recent advancements in terahertz time-domain spectroscopy (THz-TDS) have enabled sub-microgram detection and rapid spectral acquisition, making it a promising tool for automated pharmaceutical authentication. In this work, we present a THz-TDS analytical pipeline that leverages machine learning algorithms to differentiate between DHU, GA, and PIP. By employing t-distributed stochastic neighbor embedding (t-SNE) and hierarchical density-based clustering (HDBSCAN) to analyze full-spectrum multivariate patterns, we achieve a clustering accuracy of 99.38 %. This methodology offers a rapid and non-invasive approach to pharmaceutical identification, with significant implications for counterfeit detection, quality assurance, and personalized medicine. It highlights the potential of terahertz (THz) spectroscopy as a transformative tool in modern analytical chemistry.