HPLC, PTR-ToF-MS Enable AI Grading of White Tea

As sorting Baimudan white tea into different quality grades is mostly done currently subjectively by taste and smell testing from trained experts, a technique that is faster, more objective, and based on data would be preferable. With that in mind, a joint study conducted by researchers at Pusan National University (Gyeongnam, South Korea) and Fuzhou University (China) set out to test whether measuring the tea's aroma "fingerprint" using proton-transfer-reaction time-of-flight mass spectrometry (PTR-ToF-MS) combined with targeted non-volatile metabolite measurements could support data-driven grading of four official Baimudan grades (Special, Grade I, Grade II, and Grade III), with non-volatile metabolites quantified by high-performance liquid chromatography (HPLC), amino acid analysis, and soluble sugar assay. A paper based on their work was published in the journal Rapid Communications in Mass Spectrometry.¹

What Determines Tea Quality/Grade, and Why Is a New Approach Needed?

How good a tea tastes and smells come down to its chemical makeup and how it was processed. Baimudan white tea specifically is officially sorted into four abovementioned quality tiers based on a national grading standard.^2,3Earlier research has already shown that tea samples from different quality grades tend to differ in both their aroma makeup and their underlying chemical composition.^4-6Trained taste-testing panels are still the go-to method for grading tea in practice, but this approach can take a lot of time, and results can shift depending on how experienced the panel members are or the exact conditions the tea is tested under.⁷ These shortcomings, in the opinion of the authors of the paper,¹ motivate “complementary, instrument-based approaches that quantify grade-relevant chemical signatures to support standardized quality control.”

How Well Did the Model Perform at Grading Tea, and What Patterns Did It Rely on to Make Its Decisions?

The researchers tested a total of 120 tea samples (30 from each of the four grades) splitting them into a larger group used to train an AI model and a smaller separate group used to check how well it worked. They built an AI system that separately analyzed two types of data (the tea's aroma profile and its chemical composition) before combining them to sort each sample into the correct grade. When tested on the separate "unseen" group of samples, the AI correctly identified the grade 23 out of 24 times (about 96% accuracy) with the only mistake being a mix-up between two grades that were very close in quality. The researchers double-checked their results using multiple rounds of testing on different portions of the training data, and the system performed consistently well each time. When they dug into what the AI was actually picking up on, they found that higher-grade teas tended to have more floral and sweet aroma compounds, along with higher levels of amino acids and natural sugars, while lower-grade teas leaned more toward "green" or woody aroma notes and different chemical markers.¹

“Within the current dataset,” write the authors of the paper,¹ “integrating PTR-ToF-MS volatile fingerprints with targeted non-volatile metabolites provides an objective and chemically interpretable approach for instrument-assisted grading of Baimudan white tea.”

Read More on Similar Topics
AI-Powered Precision for Functional Component Testing in Tea Analysis

References

1. Zhang, D.; Hong, S. H.; Li, X. Multimodal Transformer Fusion of PTR-ToF-MS Volatiles and Targeted Non-Volatile Metabolites Enables Data-Driven Grading of Baimudan White Tea. Rapid Commun Mass Spectrom. 2026, 40 (18), e70122. DOI: 10.1002/rcm.70122
2. China SAo, GB/T 22291-2017 White Tea. State Administration for Market Regulation/Standardization Administration of China, 2017.
3. Wijesekara, T.; Liu, Y.; Xu, B. New Insights into Chemical Compositions and Health Benefits of White Tea and Development of New Products Derived from White Tea, Food Sci. Hum. Wellness 2024, 14, 9250424. DOI: 10.26599/fshw.2024.9250424
4. Yue, W.; Sun, W.; Rao, R. S. P. et al. Non-Targeted Metabolomics Reveals Distinct Chemical Compositions Among Different Grades of Bai Mudan White tea. Food Chem. 2019, 277, 289-297.DOI: 10.1016/j.foodchem.2018.10.113
5. Ning, J.-M.; Ding, D.; Song, Y.-S. et al. Chemical Constituents Analysis of White Tea of Different Qualities and Different Storage Times. Eur Food Res Technol. 2016, 242, 2093–2104. DOI: 10.1007/s00217-016-2706-0
6. Ye, J.-H.; Ye, Y.; Yin, J.-F. et al. Bitterness and Astringency of Tea Leaves and Products: Formation Mechanism and Reducing Strategies. Trends Food Sci. Tech. 2022, 123, 130–143. DOI: 10.1016/j.tifs.2022.02.031
7. Liang, Y.; J. Lu, J.; L. Zhang, L. et al. Estimation of Black Tea Quality by Analysis of Chemical Composition and Colour Difference of Tea Infusions, Food Chem. 2003, 80, 283–290. DOI: 10.1016/s0308-8146(02)00415-6