Cultivation of Cannabis for medicinal purposes primarily relies on seed propagation with expected variations in yield, cannabinoid content, growth rate and seed material non-uniformity. This study aims to employ digital methods for morphological analysis and infrared spectroscopy, combining them with multivariate analysis to characterize and differentiate Cannabis seeds. Morphological traits of 100 seeds from both commercial Cannabis specimens and wild-growing local varieties were analyzed using the high-throughput phenotyping software in addition to their collected infrared spectra in the mid-IR region. Subsequently, a chemometrics approach by means of Principal Component Analysis (PCA), Hierarchical Cluster Analysis (HCA), and Partial Least Square-Discriminatory Analysis (PLS-DA) was applied. The statistical indicators of the PLS-DA model (R2X=0.99, R2Y=0.63, Q2=0.64) demonstrate strong predictive capabilities for the differentiation of Cannabis seed specimens based on morphological attributes. The score scatter plot clearly shows a distinct grouping pattern, primarily driven by seed size. Wild-type seeds predominantly cluster into group 1, characterized by smaller diameters, while commercial seeds cluster into group 2. By analysing spectral data, in contrast to the expected differentiation based on secondary metabolites (cannabinoids) in the seeds, differentiation was based on the macronutrient profile with characteristic bands at 3275 cm, 2921 cm, 2852 cm, 1743 cm, 1630 cm, 1532 cm, 1459 cm, 1239 cm, 1157 cm, 1094 cm, 1018 cm, identified as the most distinctive spectral features. The PCA model (R2X=0.88 and Q2=0.85) was composed of 5 principal components explaining 88% of the spectral variations. The loading plot of the PLS-DA model reveals the distinctive spectral features for both groups (lipid and carbohydrate bands - group 2 samples, protein and water content - group 1 samples). The developed models have the potential for application for rapid screening of quality parameters of Cannabis seeds and their differentiation.

中文翻译：

---

采用数字形态筛选和红外光谱结合多变量建模来区分大麻种子

药用大麻种植主要依靠种子繁殖，其产量、大麻素含量、生长速度和种子材料不均匀性存在预期差异。本研究旨在采用数字方法进行形态分析和红外光谱，将其与多变量分析相结合来表征和区分大麻种子。除了在中红外区域收集的红外光谱之外，还使用高通量表型分析软件对来自商业大麻标本和野生生长的当地品种的 100 颗种子的形态特征进行了分析。随后，应用了主成分分析（PCA）、层次聚类分析（HCA）和偏最小二乘判别分析（PLS-DA）的化学计量学方法。PLS-DA模型的统计指标（R2X=0.99，R2Y=0.63，Q2=0.64）显示出基于形态属性的大麻种子标本分化的强大预测能力。分数散点图清楚地显示了明显的分组模式，主要由种子大小驱动。野生型种子主要聚类为第 1 组，其特征是直径较小，而商业种子聚类为第 2 组。通过分析光谱数据，与基于种子中次生代谢物（大麻素）的预期分化相反，分化是基于常量营养素谱的特征谱带位于 3275 cm、2921 cm、2852 cm、1743 cm、1630 cm、1532 cm、1459 cm、1239 cm、1157 cm、1094 cm、1018 cm，被确定为最独特的光谱特征。PCA 模型（R2X=0.88 和 Q2=0.85）由 5 个主成分组成，解释了 88% 的光谱变化。PLS-DA 模型的载荷图揭示了两组的独特光谱特征（脂质和碳水化合物谱带 - 第 2 组样品，蛋白质和水含量 - 第 1 组样品）。开发的模型具有用于快速筛选大麻种子质量参数及其区分的潜力。