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

Thepin N.; Somboonsaksri P.; Lapkrajang D.; Termsaithong N.; Limwichean S.; Jintamethasawat R.; Botta R.; Ngernsutivorakul T. · Talanta · 2026

**Background & Methods** This review describes development of an improved surface-enhanced Raman spectroscopy (SERS) detection method for Δ9-tetrahydrocannabinol (THC) using chemical derivatization with 4-aminophenol (4-AP) and silver nanorod substrates. The method was evaluated for sensitivity, selectivity, and applicability in detecting THC in complex cannabis-infused samples and commercial beverages, with multivariate analysis and machine learning models applied for data interpretation. **Key Findings** • 4-Aminophenol derivatization significantly enhanced SERS performance, achieving a 20-fold improvement in detection sensitivity compared to non-derivatized approaches, with a detection limit of 0.051 mg/L and sub-ppm performance in beverage samples without requiring sample extraction. • The method demonstrated excellent spectral differentiation of THC from other prominent cannabinoids including cannabidiol (CBD), cannabinol (CBN), and tetrahydrocannabinolic acid (THCA), with 1100 cm⁻¹ identified as the optimal THC marker band. • Machine learning classification models achieved high accuracy rates of 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, with improved performance when using THC-related spectral bands. **Dosage & Administration** Not reported. **Safety & Adverse Effects** Not reported. **Evidence Quality** This is a methodology review paper describing analytical chemistry techniques rather than a clinical trial. Evidence quality is limited by the lack of clinical validation, absence of in vivo human studies, and restriction to laboratory-based analytical performance. The study demonstrates promising technical capabilities but provides no clinical safety or efficacy data. The machine learning models, while achieving high classification accuracies, were developed on limited sample sets and would require external validation in diverse real-world cannabis products. Practical applicability is further constrained by the requirement for specialized SERS equipment and technical expertise. The work represents important advancement in forensic and quality control methodologies but cannot inform clinical cannabis use or therapeutic applications.

More recent cannabis research