目的 构建通关藤真伪的识别智能模型，并探讨通关藤的“辨味论质”内在机制。方法 采用电子舌技 术测定 24 批通关藤及其 4 批混淆品味道的响应值，并利用主成分分析（PCA）、偏最小二乘判别分析（PLS-DA） 方法构建通关藤真伪识别智能模型，筛选影响通关藤质量的差异味道。采用机器学习 Bittersweet 平台预测和电 子舌验证的方法筛选通关藤苦味-药效相关成分，并采用紫外-可见分光光度法（UV-Vis）、高效液相色谱法 （HPLC）测定其苦味-药效相关成分含量；MTT 法检测 24 批通关藤在 HGC-27 胃癌和 SMMC-7721 肝癌细胞的 生长抑制率，并采用相关性分析方法分析通关藤成分（通关藤苷 I、G 和总皂苷）含量和肿瘤细胞生长抑制率与 药材苦味响应值的相关性。结果 通关藤及其混淆品的味道差异明显，其中 4 种混淆品苦味和苦味回味响应值 均明显大于正品。建立的基于味道的 PCA、PLS-DA 模型可将正品和混淆品明显分成两个区域，有效区分通关 藤的真伪，并且苦味是影响通关藤质量的最大差异味道。机器学习 Bittersweet 平台预测得到 28 个苦味成分， 均为甾体皂苷类，其中通关藤苷 I、G 和总皂苷为苦味-药效相关成分。24 批通关藤中通关藤苷 I、通关藤苷 G、总皂苷成分含量范围分别为 0.18～5.34、0.57～3.43 mg·g-1和 4.24%～7.51%；24 批通关藤在 HGC-27 胃癌 和 SMMC-7721 肝癌细胞的生长抑制率范围分别为 32.07%～88.38%、32.00%～76.78%。相关性分析结果显示， 通关藤 3 种苦味-药效相关成分（通关藤苷 I、G 和总皂苷）含量、抑制肿瘤细胞生长抑制率均与药材的苦味响应 值呈显著正相关（P＜0.05）。结论 该研究所构建的基于“味”的定性识别模型可快速鉴别通关藤的真伪，其 “辨味论质”机制可能与通关藤苷 I、G 和总皂苷有关，可为通关藤中药质量评价及药效研究提供新思路。

Objective To construct an intelligent model for authenticity identification of Marsdeniae Tenacissimae Caulis and explore the intrinsic mechanism of its “taste-based quality discrimination”. Methods An electronic tongue was used to measure taste response values of 24 batches of authentic Marsdeniae Tenacissimae Caulis and four batches of its adulterants. Principal component analysis （PCA） and partial least squares-discriminant analysis （PLS-DA） were employed to build an intelligent model for authenticity identification and screen for taste attributes affecting quality differences. The Bittersweet machine learning platform and electronic tongue validation were used to screen bitter tasteefficacy related components in Marsdeniae Tenacissimae Caulis. The contents of these components were determined using Ultraviolet-Visible spectrophotometry （UV-Vis） and High Performance Liquid Chromatography （HPLC）. The inhibitory effects of 24 batches of Marsdeniae Tenacissimae Caulis on the growth of HGC-27 gastric cancer cells and SMMC-7721 liver cancer cells were detected via the MTT assay. Correlation analysis was conducted to examine the relationships between the contents of components （Tenacissoside I，Tenacissoside G，and total saponins），tumor cell growth inhibition rates， and the bitterness response value of the medicinal material. Results Significant taste differences were observed between authentic Marsdeniae Tenacissimae Caulis and its adulterants. The four adulterants exhibited markedly higher values for bitterness and bitter aftertaste compared to the authentic samples. The established taste-based PCA and PLS-DA models distinctly separated authentic and adulterant samples into two regions， effectively differentiating their authenticity， with bitterness identified as the primary taste attribute affecting quality variation. The Bittersweet platform predicted 28 bitter components，all belonging to steroidal saponins，among which Tenacissoside I，Tenacissoside G，and total saponins were identified as the key bitter taste-efficacy related components. In the 24 authentic batches，the contents of Tenacissoside I，Tenacissoside G，and total saponins ranged from 0.18 to 5.34 mg·g-1 ，0.57 to 3.43 mg·g-1 ，and 4.24% to 7.51%，respectively. The growth inhibition rates on HGC-27 and SMMC-7721 cells ranged from 32.07% to 88.38% and 32.00% to 76.78%，respectively. Correlation analysis revealed significant positive correlations （P＜0.05） between the three bitter taste-efficacy components （Tenacissoside I， Tenacissoside G， and total saponins） and their tumor cell growth inhibition rates， as well as the bitterness of the medicinal material. Conclusion The qualitative identification model based on “taste” constructed in this study can rapidly distinguish authentic Marsdeniae Tenacissimae Caulis from adulterants. Its “taste-based quality discrimination” mechanism is related to Tenacissoside I，Tenacissoside G，and total saponins，providing a new approach for quality evaluation and efficacy research of Marsdeniae Tenacissimae Caulis in Chinese medicinal materials.

R284.1

国家自然科学基金项目（82173935）；辽宁省教育厅面上项目（LJKMZ20221312）；辽宁省自然基金博士启动项目（2024-BS-134）； 辽宁省教育厅重点攻关项目（JYTZD2023198）。