Improved SERS detection of tetrahydrocannabinol using 4-aminophenol derivatization and machine learning

Fast, sensitive, and selective detection of Δ⁹-tetrahydrocannabinol (THC) is essential for forensic screening and quality control of cannabis-derived products. Although surface-enhanced Raman spectroscopy (SERS) is promising for THC detection, its practical use remains limited by insufficient sensitivity and selectivity, especially in complex cannabis-infused samples. In this work, chemical derivatization was used to enhance SERS performance for THC detection. Two commonly used colorimetric reagents, i.e., Fast Blue BB (FBBB) and 4-aminophenol (4-AP), were evaluated with silver nanorod SERS substrates. Among the tested reagents, 4-AP showed better compatibility with SERS, yielding clear and enhanced Raman fingerprints at multiple bands, with 1100 cm^-1 as the most suitable THC marker. Compared with our previous non-derivatized approach, the developed method lowered the detection limit by 20-fold, reaching 0.051 mg/L. The method also enabled spectral differentiation of THC in the presence of other prominent cannabinoids, including cannabidiol (CBD), cannabinol (CBN) and tetrahydrocannabinolic acid (THCA). Its applicability was further demonstrated by detecting THC in commercial cannabis-infused beverage samples, achieving a sub-ppm detection limit without sample extraction. Multivariate PCA and supervised machine learning models were further developed to support data analysis. Using selected THC-related bands improved chemical interpretability and provided more consistent performance across the evaluated models. Classification accuracies were 92.5-98.7% for discrimination between THC-containing and THC-free samples and 84.8-88.4% for threshold-based screening of spiked beverage samples. Overall, combining simple chemical derivatization with SERS and machine learning provides a practical approach for sensitive and selective THC analysis, with potential for screening cannabis-infused products.