E A Souza Thaineh, Bertol Gustavo, M Santos Poliana
Universidade Tecnológica Federal do Paraná, 81280-340 Curitiba, Paraná, Brazil.
Dall PhytoLab SA, 82540-040 Curitiba, Paraná, Brazil.
ACS Omega. 2025 Apr 8;10(15):15143-15147. doi: 10.1021/acsomega.4c10753. eCollection 2025 Apr 22.
This study presents a novel analytical approach for classifying commercial hemp oil samples according to their Δ9-tetrahydrocannabinol (THC) content, employing mid-infrared (MIR) spectroscopy combined with machine learning algorithms. A total of 204 commercial hemp oil samples, with THC concentrations ranging from 0.0% to 16.6% w/w, were analyzed. Partial least-squares-discriminant analysis (PLS-DA) was employed for classification purposes. Two classification models were developed based on international regulatory thresholds: model A, which classifies samples with THC concentrations exceeding 0.2% w/w, and model B, designed to classify those with THC levels above 0.3% w/w. Both models demonstrated good performance, achieving accuracy values higher than 88.50%. Notably, model B reduced false negatives, improving sensitivity (STR) values from 93.75% to 98.31% for the training set and from 77.27% to 95.00% for the test set, compared to model A. This approach offers a viable alternative to conventional laboratory methods by eliminating complex sample preparation steps and enabling simple and rapid THC screening.
本研究提出了一种新颖的分析方法，用于根据Δ9-四氢大麻酚（THC）含量对商业大麻油样品进行分类，该方法采用中红外（MIR）光谱结合机器学习算法。共分析了204个商业大麻油样品，其THC浓度范围为0.0%至16.6%（w/w）。采用偏最小二乘判别分析（PLS-DA）进行分类。基于国际监管阈值开发了两个分类模型：模型A，用于对THC浓度超过0.2%（w/w）的样品进行分类；模型B，旨在对THC水平高于0.3%（w/w）的样品进行分类。两个模型均表现出良好的性能，准确率均高于88.50%。值得注意的是，与模型A相比，模型B减少了假阴性，训练集的灵敏度（STR）值从93.75%提高到98.31%，测试集的灵敏度值从77.27%提高到95.00%。这种方法通过省去复杂的样品制备步骤，实现简单快速的THC筛查，为传统实验室方法提供了可行的替代方案。
