Optimization and flavor characterization of milk tea formulated with large-leaf yellow tea and tangerine peel: A combined fuzzy mathematical, response surface, and multimodal sensory approach Authors: Wei Zhao, Zhenbin Chen, Yanqun Jiang, Caiyan Sheng, Mingxia Lu, Jixin Zhang, Chunzhao Gong, Chenyang Shao, Jingming Ning Published: 2026 ABSTRACT This study optimized a milk tea formulated with large-leaf yellow tea (LYT) and tangerine peel and explored its flavor characteristics and potential flavor contributors. The optimal formulation (OPF) was obtained using a fuzzy mathematical response surface model. E-nose analysis showed that the OPF exhibited a well-balanced roasted, fruity, and milky aroma profile. Sensomics analysis identified six important volatiles (OAV > 1) that may be responsible for the dominant woody, fruity and roasted aromas. Ingredient contribution analysis indicated that LYT powder primarily provided roasted aroma, tangerine peel powder contributed fruity and woody aromas. E-tongue and computer vision analyses indicated that the OPF had a rich taste profile and balanced color characteristics. Moderate catechin retention reconciled functional properties with sensory palatability. Consumer validation verified high acceptance for the OPF. These findings link formulation parameters to perceptual and chemical drivers, supporting value-added utilization of LYT in beverages. Introduction Large-leaf yellow tea (LYT) is a traditional Chinese tea with a long history. It is known for its mellow taste, distinctive roasted aroma, appealing flavor attributes, and potential health benefits (Li et al., 2024). A previous study reported that, among milk teas prepared with black tea, dark tea, oolong tea, flower tea, green tea, and LYT, the LYT-based milk tea showed a mellow and rich flavor, suggesting that LYT may be a promising ingredient for milk tea development (Yang et al., 2022). Although LYT has a relatively high yield, its added value remains relatively low. Moreover, tangerine peel provides a mildly bitter and citrus-like aroma. Milk tea formulated with LYT exhibits a pleasant flavor, and the incorporation of tangerine peel into an LYT milk tea matrix creates a more complex flavor profile (Xin et al., 2018). Recently, the innovative blending of tea with diverse ingredients has emerged as a promising strategy. This approach has consequently become a focal point in contemporary tea research (Yang et al., 2022). However, systematic studies on formulation optimization, flavor characteristics, flavor formation mechanisms, and consumer acceptance of LYT milk tea remain limited. In-depth investigation of product development, evaluation models, and the molecular mechanisms underlying the flavor of LYT milk tea is of critical importance for facilitating the development of the tea industry. Therefore, we developed a milk tea formulated with tangerine peel and LYT (TALMT) using a fuzzy mathematical sensory/response surface model. Moreover, we analyzed and characterized its flavor profile, flavor sources, key flavor factors, and acceptance through multimodal sensory approaches and metabolomics (Liu et al., 2024). The combination of fuzzy mathematical sensory evaluation and response surface methodology (RSM) provides an effective approach for optimizing milk tea formulations (Yang et al., 2022). Based on sensory evaluation, the fuzzy mathematical sensory evaluation model applies fuzzy logic to comprehensively integrate multiple sensory factors, thereby more accurately reflecting the overall flavor and quality of milk tea (Xu et al., 2025). Among the key sensory attributes of milk tea, aroma is particularly important, as it plays a major role in shaping flavor perception and consumer acceptance; meanwhile, taste and color are also important components of overall product quality (Yang et al., 2025; Zhai et al., 2022). By mimicking the human olfactory system, the E-nose enables the objective quantification of the aroma quality of milk tea. Headspace solid-phase microextraction (HS-SPME) coupled with gas chromatography–mass spectrometry (GC–MS) is widely used to extract, identify, and quantify volatile compounds in tea beverages (Huang et al., 2024). E-nose can quickly characterize differences in the aroma profiles of samples, while GC–MS can further analyze the composition of their volatile compounds; the two methods complement each other to jointly reveal the aromatic differences between samples and their underlying causes. E-tongue and computer-vision technology are commonly used methods for quantifying the taste and color quality of milk tea (Mao et al., 2021). Finally, consumer validation, generally conducted using Likert scales, is essential for evaluating market acceptance of milk tea and guiding subsequent flavor optimization (Martínez-Mendoza et al., 2020). In this study, we developed a milk tea formulated with tangerine peel and LYT using a fuzzy mathematical sensory/response surface model. Based on this, we analyzed and characterized its flavor profile, ingredient contribution, and important flavor factors through multimodal sensory approaches and metabolomics. Finally, the acceptance of the product was investigated through interdisciplinary methods combining sensory science and consumer validation. This study systematically integrated formulation optimization, sensory modeling, instrumental analysis, and consumer validation into the development and evaluation of a novel milk tea system prepared with LYT and tangerine peel. This study may provide valuable insights and a technical framework for future tea beverage development. Materials and methods Samples and sensory evaluation The tea samples used in the experiment were LYT Special First-Class (Huibinyi Tea Co., Ltd., China), 5-year-old Xinhui tangerine peel (Bozhou, China), milk powder (MP; Nestlé Co., Ltd., China), and Siraitia grosvenorii zero-calorie sweetener (SGZCS) (Huacheng Bioresources Co., Ltd., China). The same batch of LYT and tangerine peel was crushed, extracted, filtered, sterilized, concentrated, and freeze-dried to obtain the corresponding large-leaf yellow tea instant powder (LYTIP) and tangerine peel instant powder (TPIP) (Fig. S1). Table S1 presents the basic information of the reagents used. All aroma standards had a purity >95%. All sensory evaluation procedures were in accordance with the regulations and approved by the Science and Technology Ethics Committee of Anhui Agricultural University (approval number: KJLL2025069). Written informed consent was obtained from all participants prior to the sensory evaluation, including consent to participate and consent for the use of their anonymized data for research and publication. The rights and privacy of the participants were protected throughout the research; no personal identifying information was collected or disclosed. A total of 10 experienced panelists (4 men and 6 women, aged 22–40) who had sufficient knowledge about the quality assessment of milk tea participated in the sensory evaluation. All the panelists had at least 4 years of sensory evaluation experience to ensure the accuracy of the sensory evaluation results. To fully reflect the flavor characteristics of the ingredients and avoid the reduction of milk's nutritional value due to high temperatures, we set the evaluation brewing temperature at 80 °C ± 5 °C. Specifically, the different ingredients and formulations were added to 100 mL of water at 80 ± 5 °C in the sensory evaluation room. The samples were steeped for 5 min until fully dissolved, and the panelists then performed a blind evaluation of the sensory attributes. Between samples, panelists were required to rinse their mouths with purified water to cleanse their palates (Xu et al., 2025). Acquisition and optimization of formulation for TALMT Single-factor test The fuzzy mathematical evaluation model established for formulating TALMT is as follows. The factor set of TALMT was defined as V1: excellent (100–90 points), V2: good (89–80 points), V3: fair (79–70 points), and V4: poor (< 70 points). To determine the weights of the evaluation indices, the 10 trained panelists rated the importance of each index based on review requirements and relevant literature (Xu et al., 2025). The weights assigned to color, aroma, taste, appearance and texture, and overall harmony were 0.15, 0.15, 0.50, 0.10, and 0.10, respectively. The weight vector (A) of each factor was 0.15, 0.15, 0.50, 0.10, and 0.10 for A = a1, a2, a3, a4, and a5, respectively. The comprehensive sensory score was calculated using the weighted average method based on the membership degree of each evaluation grade. The specific sensory scoring criteria of milk tea are presented in Table S2 (Bao, 2025; Xu et al., 2025). Based on preliminary experiments, five addition levels were set for each ingredient, namely LYTIP (0.10, 0.30, 0.50, 0.70, and 0.90 g), TPIP (0.10, 0.20, 0.30, 0.40, and 0.50 g), MP (2, 4, 6, 8, and 10 g), and SGZCS (1, 2, 3, 4, and 5 g). Specifically, LYTIP, TPIP, MP, and SGZCS were added to 100 mL of water at 80 °C ± 5 °C with the predetermined addition gradients; the fuzzy mathematical sensory score was reviewed and calculated by the panelists. RSM modeling of making TALMT Based on the results of the single-factor experiments, a response surface optimization test was conducted using four factors as independent variables and the fuzzy mathematical sensory score as the response variable (Table S3). The formulation with the highest score and the one with the second-highest score in the fuzzy sensory evaluation were defined as formulation one (PF1) and formulation two (PF2). A mathematical model was established to calculate the optimal formulation (OPF) using Design-Expert. Basic physicochemical and hygienic index detection methods for TALMT A pH meter (PHS-25) was used to determine the pH value of the samples. The sugar content was measured using a brix refractometer (Zhejiang Top Instrument Co., Ltd., China). The protein content was determined in accordance with GB 5009.5–2016. Catechins, caffeine, and tea polyphenols were determined according to GB/T 8313–2018. The total colony count, and coliform population were measured using GB 4789.2–2022 and GB 4789.3–2016, respectively. Analysis of sample aroma quality The E-nose (AIRSENSE, PEN3) method was slightly modified based on previous research (Liu et al., 2022). Three formulations (PF1, PF2, and OPF) and four ingredients (LYTIP, TPIP, MP, and SGZCS) were added to the headspace bottle and brewed with 100 mL of water (80 °C ± 5 °C). The brewing method was consistent with that described in Section 2.1. The mixture was sealed with double-layer film and placed in a water bath for 5 min. An E-nose was used to collect odor fingerprints. The operating parameters of the E-nose included a sensor cleaning time of 120 s, a zeroing time of 5 s, a sample preparation time of 5 s, and a sample collection time of 60 s. The performance of each sensor in the E-nose is detailed in Table S4. Volatiles were extracted using a modified method based on Yang et al. (2025). Briefly, 10 mL of the test sample was transferred into a 20-mL headspace vial. Subsequently, 3 g of NaCl and 5 μL (30 μg/mL) of an internal standard solution of ethyl decanoate were added, and the mixture was thoroughly mixed. The vial was placed in a constant-temperature water bath at 60 °C for 5 min to reach equilibrium. A 2-cm extraction head (50/30-μm divinylbenzene/carboxyphene/polydimethylsiloxane-coated fiber) was inserted into the headspace vial, which was positioned 1 cm above the liquid level, and it was allowed to adsorb for 30 min for subsequent GC–MS analysis. GC conditions: a DB-5MS capillary column (30 m × 0.25 mm × 0.25 μm; Agilent J&W Scientific, Folsom, CA, USA) was used. High-purity helium gas was used as a carrier (>99.999%) at a constant flow rate of 1 mL/min. The inlet temperature was set to 250 °C, and the samples were injected in non-split mode. The oven program was set as follows: hold at 50 °C for 2 min, increase to 100 °C at 10 °C/min and hold for 0 min, increase to 160 °C and hold for 0 min, and lastly, increase to 280 °C at 25 °C/min and hold for 3 min. MS conditions were set as follows: an electron impact ion source was used with an ion source temperature of 230 °C, a quadrupole temperature of 150 °C, and an interface temperature of 280 °C. electron energy was 70 eV. Data were acquired in full-scan mode over a mass range of m/z 30–350. Qualitative and quantitative analyses of volatiles: the retention index (RI) of volatiles was calculated using the retention time of normal alkanes (C6–C21) as performed for reference compounds. The RI values were compared with those reported by the National Institute of Standards and Technology and in the literature, and compounds with RI differences of >20 were excluded (Chen et al., 2025). Ethyl decanoate was used as an internal standard for relative quantification of volatiles. Quantification was based on the following principle: the ratio of the peak area of the internal standard to the peak area of the volatile compound is equal to the ratio of the concentration of the internal standard to the concentration of the volatile compound. The odor activity value (OAV) is a common indicator used to assess the importance of aroma compounds. The calculation formula used was as follows: OAV = C/OT, where C represents the absolute concentration of the volatile (μg/kg) and OT denotes its odor threshold (OT) in water (μg/kg). Analysis of sample taste and color quality E-tongue (SA402B, INSENT, Japan) quantification was performed to detect and measure bitterness, astringency, sourness, umami, and sweetness and to analyze the differences in taste quality among samples (Peris & Escuder-Gilabert, 2016). After brewing (Section 2.1), the samples were cooled to 25 °C. A 35-mL aliquot of the sample was transferred to a specialized E-tongue measuring cup and placed in the E-tongue instrument for taste data collection. The cleaning time for the positive and negative electrodes of the system was 90 s. The cleaning time for the reference solution was 120 s for each of the two washes. The sensor operation time was 30 s. Each sample was measured four times. The first measurement result, which was unstable, was discarded, and the average of the remaining three measurements was used for subsequent statistical analysis. HPLC was employed to determine the concentrations of catechin and caffeine in samples. These compounds were separated using a Waters Symmetry® C18 column (150 × 3.9 mm, 5 μm). The chromatographic column was maintained at a temperature of 35 °C, a mobile phase A (acetonitrile:acetic acid:water = 9:2:89, v/v), and a mobile phase B (acetonitrile:acetic acid:water = 80:2:18, v/v), with a flow rate of 1 mL/min and an injection volume of 10 μL. Detection was performed at a wavelength of 278 nm. This method was based on the study of He et al. (2023). Absolute quantitative analysis of the detected substances was performed using a standard curve (Table S5). The methods of computer-vision technology used were based on those described by Wei et al. (2023) and Sheng et al. (2024). The digital camera (EOS 80D) was adjusted to the following parameters: no flash, ISO 400, aperture F/7.0, exposure time 1/200 s, and white balance set to manual. Subsequently, four regions of the image were selected on the computer, and the R, G, and B values of the selected areas were measured multiple times using ImageJ. Finally, the RGB image was converted into L*, a*, b*, C*, and H° via MATLAB R2016 (MathWorks, Natick, Massachusetts, United States) to represent the color of samples. In this context, L* represents the luminosity change between black (0) and white (100), a* indicates the difference between red (+) and green (−), b* indicates the difference between yellow (+) and blue (−), C* indicates the color saturation, where higher values indicate purer colors (0 = gray and 100 = most saturated), and H° indicates the hue of the color (0 = red, 180 = green, and 360 = blue). Consumer validation of TALMT The Likert scale is a widely used psychometric tool in social science research and is among the most commonly applied rating scales (Martínez-Mendoza et al., 2020). This scale consists of a series of statements, each of which is rated on a five-point scale ranging from “Strongly Agree” to “Strongly Disagree,” with corresponding scores of 5, 4, 3, 2, and 1. The total score for each participant was calculated by summing the scores of their responses to individual items, which could indicate the strength or intensity of their attitude or opinion on the product. This method is known for its objectivity and is widely applied in consumer acceptance research contexts (Lee et al., 2002). College students selected from different regions were the primary participants of the survey, with other members of the public as supplementary participants. The questionnaire was distributed online to collect responses. The questionnaire can be found at the end of the supplementary material. The product was provided for participants' tasting. This questionnaire employed a five-point Likert scale for sensory evaluation acceptance analysis. Participants rated the aroma (tangerine peel and tea), taste intensity (tea, tangerine peel, sweetness, and overall harmony), and overall acceptance of TALMT (scored 1–5). Statistical analysis All experimental data were collected in at least three replicates. Results were expressed as mean ± standard deviation. The Box–Behnken Design model was established using Design-Expert (Stat-Ease, Minneapolis, Minnesota, USA). The data were analyzed using SPSS 27 (IBM, Armonk, New York, USA) for one-way analysis of variance (ANOVA). All analyses were visually presented using Origin 2022 (OriginLab, Northampton, Massachusetts, USA) and Adobe Illustrator 2023 (Adobe Systems, San Jose, California, USA). SIMCA-P 14.1 (Umetrics, Umea, Sweden) was used to conduct multivariate statistical analysis. Results and discussion Acquisition and optimization of formulation for TALMT Single-factor fuzzy sensory evaluation LYTIP, TPIP, MP, and SGZCS make the milk tea formulation. Single-factor experiments combined with RSM were used to obtain and optimize this formulation. Sensory evaluation scores were determined according to established sensory evaluation standards (Table S6). To overcome the limitations of direct arithmetic averaging, a fuzzy comprehensive scoring method based on fuzzy mathematical principles was employed. This approach effectively alleviated the influence of human subjectivity and reduced errors in the evaluation process (Wu et al., 2022). Taking the A1 milk tea formulation from the single-factor experiment as an example, the fuzzy comprehensive evaluation was performed. According to the fundamental principles of fuzzy mathematics, the comprehensive membership of A1 for each evaluation factor (YA1) was calculated as YA1 = A × RA1. Based on the scoring data for each index, the specific calculation process for the result vector of the A1 sensory quality comprehensive evaluation is as follows:YA1=A×RA1=0.150.150.50.10.1×0.40.40.20.00.40.60.00.00.60.20.20.00.90.10.00.00.50.40.10.0=0.560.300.140.00 The comprehensive score was calculated using an evaluation formula H=∑i=1njbj combined with the weights of each index in the table. According to this formula, the fuzzy mathematical values for each sensory attribute were obtained. As an example, for the A1 milk tea formulation, the comprehensive sensory score was HA1 = 0.56 × 90 + 0.30 × 80 + 0.14 × 70 + 0.00 × 60 = 84.20. Other formulations were evaluated using the same methodology. The resultant single-factor sensory scores under varying raw material concentrations are shown in Fig. S2. Fuzzy comprehensive sensory evaluation indicated that the overall sensory score exhibited a typical trend of first increasing and then decreasing with increasing addition level. Accordingly, three consecutive levels around the peak value were selected as the low, medium, and high levels for subsequent single-factor experiments: LYTIP at 0.30, 0.50, and 0.70 g; TPIP at 0.10, 0.20, and 0.30 g; MP at 4, 6, and 8 g; and SGZCS at 2, 3, and 4 g. RSM modeling Based on the results of single-factor experiments (Fig. S2), the raw material parameters of the TALMT were optimized. To further investigate the relationship between the number of ingredients added and the sensory score of TALMT, the Box–Behnken Design model was established (Table S7). To test the reliability and predictive ability of the response surface model, an ANOVA analysis was conducted on the model. The model exhibited a high F-value (10.27) with a corresponding p-value <0.0001, indicating that the model was statistically significant. The p-value for Lack of Fit is >0.05, suggesting that the model has adequately fitted the data and that no key factors have been omitted. Furthermore, R2Pre (0.91) and R2Adj (0.81) were relatively high, with a difference of <0.20, indicating that the model fit well (Table S8) (Liu et al., 2024). With other factors held constant, the impact of individual factors on sensory quality was visualized using a response surface diagram (Fig. 1). Comparing the six contour plots, the significance of the additive interactions was as follows: AB (LYTIP and TPIP) > AD (LYTIP and SGZCS) > AC (LYTIP and MP) > BD (TPIP and SGZCS) > BC (TPIP and MP) > CD (MP and SGZCS). The predicted sensory score was estimated to be 84.919 points using the Design-Expert 13.0 software fitting. The OPF was determined to be as follows: LYTIP (0.676 g), TPIP (0.230 g), MP (6.269 g), and SGZCS (3.309 g). Under optimized conditions, the actual score of the OPF for milk tea was 85.75, which was considerably close to the predicted value of 84.92 from the response surface model. The result indicated good consistency between the experimental results and model predictions (Yang et al., 2022). Lastly, the response surface model generated the formulation parameters of the OPF based on the formula parameters and the fuzzy mathematical sensory evaluation (FMSE) scores: LYTIP (0.70 g), TPIP (0.20 g), MP (6.30 g), and SGZCS (3.30 g).Fig. 1Response surface of the additives amount to sensory evaluation. Note: Fuzzy mathematical sensory score (FMSE), large-leaf yellow tea instant powder (LYTIP), tangerine peel instant powder (TPIP), Siraitia grosvenorii zero-calorie sweetener (SGZCS), milk powder (MP). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)Fig. 1 Response surface of the additives amount to sensory evaluation. Note: Fuzzy mathematical sensory score (FMSE), large-leaf yellow tea instant powder (LYTIP), tangerine peel instant powder (TPIP), Siraitia grosvenorii zero-calorie sweetener (SGZCS), milk powder (MP). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) To conduct a more thorough comparative analysis of the flavor of TALMT, in the subsequent experiments, we compared the two formulations—PF1 and PF2 (the top two in the fuzzy sensory score ranking)—with the best formulation—OPF (the one optimized by the model). Basic physicochemical and hygienic index analyses for different formulations and ingredients At present, low-sugar milk tea is a primary direction for future tea development. SGZCS is widely used in the beverage industry owing to its low-calorie, high-sweetness, non-fermentation, and health-promoting properties (Wang et al., 2019). The total sugar content of the milk tea (derived from 3 to 3.3 g/100 mL of SGZCS, Table S9) remained below the threshold of 5 g/100 mL specified in GB 28050–2011; thus, it was classified as a low-sugar milk tea. The sugar content in the three milk tea formulations ranged from 8.60 to 9.47°Bx, which aligned with current market and consumer preferences. Among these, the sugar content of OPF (9.43 ± 0.06°Bx) exhibited an appropriate sugar content in relation to overall sensory quality. The protein content of OPF was 1.66 g/100 g, with no marked differences observed among the three formulations. Notably, MP protein (1.58 ± 0.02 g/100 g) was the largest contributor to milk tea protein, followed by LYTIP (0.21 ± 0.00 g/100 g) (Table S9). Tea protein is usually insoluble in water and forms a colloidal structure in the tea infusion. Researchers have reported that tea protein contributes to the stability of tea and exhibits biological activities such as radiation prevention and lipid reduction (Qiu et al., 2023). Moreover, the pH of OPF (7.84 ± 0.04) was considerably lower than that of PF1 and PF2, indicating that a relatively lower pH may be associated with improved milk tea quality (Table S9). The TPIP in the ingredients (5.60 ± 0.02) exhibited the lowest pH, showing weak acidity and considerably contributing to the reduction of milk tea pH. Product safety is a fundamental prerequisite for research on milk tea. To determine whether the authenticity, safety, and health risk indicators of the best-quality OPF comply with national standards, we tested the tea polyphenols, caffeine, total colony count, and coliform population of OPF. The results indicate that all indicators meet the quality standards requirements specified in the standard (Table S10). This confirmed that the materials used for the best-quality OPF milk tea were genuine and reasonable, and the standards were safe. Analysis of aroma quality of samples Aroma profile difference analysis The aroma of milk tea is a crucial factor in determining its quality, as the aromatic compounds directly reflect the quality characteristics of the product and considerably influence consumer preferences (Yang et al., 2025; Zhai et al., 2022). To evaluate the aroma profiles of the OPF, PF1, and PF2, as well as to elucidate the differences among them and the contributions of their key ingredients, we conducted E-nose on the formulations and their individual ingredients. Through the E-nose, the aroma characteristics of three milk tea formulations (PF1, PF2, and OPF) and their ingredients (LYTIP, TPIP, MP, and SGZCS) were intuitively compared and analyzed (Fig. 2). The response values of the E-nose sensors between the three formulations (PF1, PF2, and OPF) considerably differed (Fig. 2A). Among the 10 sensors, the response values of sensors S9 (W2W) and S10 (W3S) markedly differed across various formulations (p < 0.05). The response of OPF on S9 (W2W) and S10 (W3S) was lower than that of the other two formulations (p < 0.05). Sensors S9 (W2W) and S10 (W3S) were observed to be sensitive to aromatic, sulfur-containing compounds and alkanes, respectively, indicating that the content of aromatic, sulfur-containing compounds and alkanes in the OPF was low (Li et al., 2024). Excessively high levels of these aroma-active substances could cause masking or conflicts among aroma notes, thereby hindering the formation of a balanced and harmonious aroma profile. The aroma intensity of OPF was moderate, avoiding such masking and conflicts, resulting in a balanced and harmonious aroma profile (Yang et al., 2022). The response values of the remaining sensors showed no significant differences (p > 0.05).Fig. 2Quantitative analysis of different formulations and ingredients flavor. Note: (A-B) Aroma profile of E-nose in formulations and ingredients; (C—D) Taste profile of E-tongue in formulations and ingredients, two better formulations (PF1, PF2), optimal formulation (OPF), large-leaf yellow tea instant powder (LYTIP), tangerine peel instant powder (TPIP), Siraitia grosvenorii zero-calorie sweetener (SGZCS), milk powder (MP). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)Fig. 2 Quantitative analysis of different formulations and ingredients flavor. Note: (A-B) Aroma profile of E-nose in formulations and ingredients; (C—D) Taste profile of E-tongue in formulations and ingredients, two better formulations (PF1, PF2), optimal formulation (OPF), large-leaf yellow tea instant powder (LYTIP), tangerine peel instant powder (TPIP), Siraitia grosvenorii zero-calorie sweetener (SGZCS), milk powder (MP). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) Marked differences were observed among the ingredients (LYTIP, TPIP, MP, and SGZCS) (Fig. 2B). The response values of each raw material on the sensor S7 were the largest, and there were notable differences. Among the raw materials, LYTIP showed the highest response value in S7, followed by TPIP. S7 was highly sensitive to sulfur-containing compounds and heterocyclic compounds. Studies demonstrated that heterocyclic compounds were the primary contributors of the aroma of LYT, which was consistent with the E-nose detection results observed in this study (Guo et al., 2021). Overall, LYTIP and TPIP showed stronger sensor responses and thus contributed more prominently to the aroma of the milk tea, while SGZCS exhibited low response values and a correspondingly low aroma contribution. In summary, there were notable differences in the response values of E-nose sensors between PF1, PF2, and OPF. The OPF maintained a moderate content of aromatic, sulfur-containing, and alkylated compounds, ensuring a balanced and harmonious aroma profile. This was one of the reasons that contributed to the superior aroma quality of OPF. Among the ingredients, LYTIP and TPIP contributed most strongly to the aroma, whereas SGZCS and MP contributed the least. Analysis of volatile compounds in milk tea Volatile compounds identification To clarify the differences in aroma among the three formulations of TALMT and to understand the chemical basis underlying the OPF, HS-SPME–GC–MS was employed to profile and quantify their volatiles. Fifty-four volatiles were detected across TALMT, such as heterocyclics (12), alcohols (11), aldehydes (10), ketones (10), esters (5), aromatics (2), alkenes (2), acids (1), and others (1). The volatiles present in higher concentration were hydrocarbons, alcohols, aldehydes, ketones, and esters, which together constituted the aroma profile and differences among the different formulations (Table S11). To intuitively analyze the differences in the volatile content distribution of each formula milk tea, the data were visualized using heat maps (Fig. 3). The results demonstrated that OPF was primarily composed of ketones, heterocyclics, and alcohols. Notably, the concentrations of 2-heptanone and 2-nonanone in OPF were higher than those in PF1 and PF2. These ketones generally impart fruity and milky aromas, which can contribute to the milky and fruity aroma of OPF (Zhai et al., 2022). Heterocyclic compounds, particularly pyrazines, were the key contributors of roasted aroma (Zhai et al., 2022). PF1 exhibited the lowest levels of heterocyclics, such as 2,5-dimethylpyrazine, leading to a relatively weak roasted note. By contrast, PF2 contained the highest levels of heterocyclics, such as 2-ethyl-3-methylpyrazine, which could result in an overly intense roasted aroma. The balanced composition of OPF mitigated the deficiencies observed in PF1 and PF2. The concentration of heterocyclic compounds was moderate, there was no excessive masking or insufficient strength, and the aroma sensory quality was optimal.Fig. 3Single-omics correlation analysis of color values and FMSE of TALMT. Note: Fuzzy mathematical sensory score (FMSE), tangerine peel–large-leaf yellow tea milk tea (TALMT). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)Fig. 3 Single-omics correlation analysis of color values and FMSE of TALMT. Note: Fuzzy mathematical sensory score (FMSE), tangerine peel–large-leaf yellow tea milk tea (TALMT). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) Alcohols played an important role in shaping the aromatic profile and formulation differences of milk tea. OPF contained higher levels of 1-hexanol, linalool, α-terpineol, and benzyl alcohol than PF1 and PF2. Of these, linalool and benzyl alcohol imparted floral and citrus notes, enhancing the overall floral and fruity aroma of OPF (Zhai et al., 2022). α-Terpineol imparted woody and citrus-like aromas, enhancing these notes in OPF (Sales et al., 2020). The concentration of terpinen-4-ol in OPF was markedly higher than in PF1. This compound exhibited a distinct herbal aroma, which could interact with the pronounced roasted aroma of LYT, thereby enriching the overall aroma profile of milk tea (Wang et al., 2022). Among aldehydes, benzaldehyde was found at relatively high levels in OPF, contributing bitter-almond and nutty aromas (Zhai et al., 2022). In addition, hexanal, nonanal, decanal, and safranal are commonly associated with grassy, fruity, and medicinal aromas (Tian et al., 2024). In the OPF, the concentrations of these aldehydes were moderate, and they contributed to fresh and pleasant top note aromas. Notably, decanal was the most abundant aldehyde in OPF, which contributed to a more pronounced citrus-like aroma (Wang et al., 2024). Overall, the three formulations exhibited distinct volatile profiles. OPF contained moderate and well-balanced levels of pyrazines, alcohols, and aldehydes, resulting in harmonious roasted, fruity, woody, herbal, and medicinal notes. This balanced composition prevented any single aroma from becoming dominant, thereby contributing to the superior overall aroma quality of OPF. Screening of key differential volatiles This study performed multivariate statistical analyses of the peak areas of 54 identified volatiles using PCA, HCA, and PLS-DA to screen key differential volatiles and further elucidate the differences among formulations (Fig. 4) (Zhang et al., 2023). The PCA score plot showed that PF1 was in the third and fourth quadrants, PF2 was primarily distributed in the second quadrant, and OPF clustered in the first quadrant (Fig. 4A), indicating pronounced differences in the volatile metabolite profiles. These findings were consistent with the clustering patterns observed in HCA (Fig. 4B). PLS-DA was subsequently conducted (Fig. 4C), followed by model validation using permutation testing. Permutation results (R2 = 0.416, Q2 = −0.203) confirmed that the model was not overfitted and was therefore reliable for distinguishing volatile profiles among the formulations (Fig. 4D) (Hua et al., 2024).Fig. 4Analysis of volatiles in different formulations. Note: Two better formulations (PF1, PF2), optimal formulation (OPF).Fig. 4 Analysis of volatiles in different formulations. Note: Two better formulations (PF1, PF2), optimal formulation (OPF). Variable importance in projection (VIP) represents the contribution rate of each variable to the PLS-DA model. The larger the VIP value of a general compound, the higher the contribution rate to the difference model. p < 0.05 was considered statistically significant. To identify volatiles that could contribute essentially to the quality differences among milk tea formulations, 25 differential volatiles with VIP > 1 were screened (Fig. S3). To further enhance the accuracy of the screening process, VIP > 1 and p < 0.05 were used as the screening criteria, and finally 10 key differential volatiles were screened: 2-methylbutanal, 2-ethyl-6-methyl-pyrazine, linalool, p-cymene, isoborneol, p-cymen-8-ol, α-terpineol, decanal, safranal, and perillaldehyde. Therefore, these 10 key differential volatiles were identified as the important contributors to the aromatic differences in different formulations. The OAV was used to assess the sensory contribution of volatiles. Compounds with OAV > 1 were generally considered to substantially contribute to aroma (Ye et al., 2024; Zhu et al., 2021). To further evaluate the contribution of the 10 key differential volatiles to the aroma of milk tea, external calibration quantitative analysis was performed, and their OAV values were calculated (linear equations are provided in Table S12). A higher OAV value indicates a higher contribution of the volatile compound to the aroma profile of the sample (Zhang et al., 2022). Results (Table 1) indicated that in OPF, important volatiles with OAV > 1 include 2-methylbutanal (36.4), linalool (13.17), decanal (8.26), isoborneol (8.08), α-terpineol (7.14), and p-cymene (3.37). Among these, 2-methylbutanal and linalool have OAV > 10, indicating a marked contribution to the overall aroma of the OPF. The OAV values of aldehydes and alcohols were particularly prominent, which further enhanced the hierarchy and coordination of the aroma. Specifically, isoborneol and linalool may contribute to cool and fruity aromas, respectively, forming a balanced aroma with the MP milk aroma. Isoborneol mainly provided the cool aroma, while the floral and citrus-like aromas of linalool enhanced the richness of the aroma (Zhai et al., 2022). 2-Methylbutanal and decanal may provide sweet, roasted, and citrus-like aromas, which were highly consistent with the roasted aroma of LYT and the fruity aromas of tangerine peel. Alternatively, excessive levels of α-terpineol in PF1 could lead to overpowering woody aroma, disrupting aroma balance.Table 1Concentrations, OTs and OAVs of key differential volatiles in different formulations.No.CompoundsConcentration (μg/L) aOTs (μg/L) bOAVscPF1dPF2 dOPF dPF1 dPF2 dOPF d12-Methylbutanal17.72 ± 2.89ᶜ45.74 ± 4.03ᵇ54.60 ± 0.30ᵃ1.5 (Flaig et al., 2020)11.8230.4936.4022-Ethyl-6-methyl-pyrazine6.81 ± 0.41ᶜ14.07 ± 2.68ᵃ10.86 ± 0.29ᵇ40 (Gemert, 2003)0.170.350.273p-Cymene29.80 ± 2.30ᵃ11.49 ± 0.20ᶜ16.87 ± 1.57ᵇ5.01 (Gemert, 2003)5.952.293.374Linalool4.26 ± 0.75ᵃ0.94 ± 0.33ᶜ2.90 ± 0.82ᵇ0.22 (Gemert, 2003)19.384.2813.175Isoborneol126.20 ± 30.64ᵃ67.25 ± 4.09ᶜ98.30 ± 9.87ᵇ15 (Gemert, 2003)9.434.486.556p-Cymen-8-ol3.63 ± 0.20ᵃ1.36 ± 0.15ᶜ2.60 ± 0.30ᵇn.f–––7α-Terpineol43.91 ± 0.50ᵃ6.27 ± 2.23ᶜ24.03 ± 6.82ᵇ4.6 (Xu et al., 2023)9.551.365.228Decanal0.04 ± 0.00ᶜ0.26 ± 0.04ᵇ0.83 ± 0.07ᵃ0.1 (Gemert, 2003)0.452.598.269Safranal0.62 ± 0.01ᵃ0.10 ± 0.06ᶜ0.32 ± 0.08ᵇ3 (Guo et al., 2022)0.210.030.1110Perillaldehyde1.24 ± 0.07ᵃ0.50 ± 0.07ᶜ0.82 ± 0.09ᵇ30 (Gemert, 2003)0.040.020.03Note: Concentration a, the absolute content of the volatiles calculated based on the standard curve. OTs b, the odor threshold (OT) of volatile in water. OAV c, this is calculated by dividing the concentration of the volatile by its OT. “n.f.” indicates that the corresponding information was not found, while “--” signifies that the calculation cannot be performed due to missing data. Two better formulations (PF1, PF2), optimal formulation (OPF). Different letters in the same row indicate significant differences (p < 0.05). Table 1Concentrations, OTs and OAVs of key differential volatiles in different formulations.No.CompoundsConcentration (μg/L) aOTs (μg/L) bOAVscPF1dPF2 dOPF dPF1 dPF2 dOPF d12-Methylbutanal17.72 ± 2.89ᶜ45.74 ± 4.03ᵇ54.60 ± 0.30ᵃ1.5 (Flaig et al., 2020)11.8230.4936.4022-Ethyl-6-methyl-pyrazine6.81 ± 0.41ᶜ14.07 ± 2.68ᵃ10.86 ± 0.29ᵇ40 (Gemert, 2003)0.170.350.273p-Cymene29.80 ± 2.30ᵃ11.49 ± 0.20ᶜ16.87 ± 1.57ᵇ5.01 (Gemert, 2003)5.952.293.374Linalool4.26 ± 0.75ᵃ0.94 ± 0.33ᶜ2.90 ± 0.82ᵇ0.22 (Gemert, 2003)19.384.2813.175Isoborneol126.20 ± 30.64ᵃ67.25 ± 4.09ᶜ98.30 ± 9.87ᵇ15 (Gemert, 2003)9.434.486.556p-Cymen-8-ol3.63 ± 0.20ᵃ1.36 ± 0.15ᶜ2.60 ± 0.30ᵇn.f–––7α-Terpineol43.91 ± 0.50ᵃ6.27 ± 2.23ᶜ24.03 ± 6.82ᵇ4.6 (Xu et al., 2023)9.551.365.228Decanal0.04 ± 0.00ᶜ0.26 ± 0.04ᵇ0.83 ± 0.07ᵃ0.1 (Gemert, 2003)0.452.598.269Safranal0.62 ± 0.01ᵃ0.10 ± 0.06ᶜ0.32 ± 0.08ᵇ3 (Guo et al., 2022)0.210.030.1110Perillaldehyde1.24 ± 0.07ᵃ0.50 ± 0.07ᶜ0.82 ± 0.09ᵇ30 (Gemert, 2003)0.040.020.03Note: Concentration a, the absolute content of the volatiles calculated based on the standard curve. OTs b, the odor threshold (OT) of volatile in water. OAV c, this is calculated by dividing the concentration of the volatile by its OT. “n.f.” indicates that the corresponding information was not found, while “--” signifies that the calculation cannot be performed due to missing data. Two better formulations (PF1, PF2), optimal formulation (OPF). Different letters in the same row indicate significant differences (p < 0.05). Concentrations, OTs and OAVs of key differential volatiles in different formulations. Note: Concentration a, the absolute content of the volatiles calculated based on the standard curve. OTs b, the odor threshold (OT) of volatile in water. OAV c, this is calculated by dividing the concentration of the volatile by its OT. “n.f.” indicates that the corresponding information was not found, while “--” signifies that the calculation cannot be performed due to missing data. Two better formulations (PF1, PF2), optimal formulation (OPF). Different letters in the same row indicate significant differences (p < 0.05). Ingredients and their proportion played a crucial role in shaping the aroma and overall quality of milk tea. Volatile profiles substantially differed among LYTIP, TPIP, MP, and SGZCS, and the concentration, synergy, and masking effects of their respective volatiles jointly determined the aroma profiles of TALMT. The contribution of ingredients to the aroma quality of TALMT was analyzed based on the content of important volatiles in the ingredients (Table S13). 2-Methylbutanal, which is associated with roasted aroma, was detected only in LYTIP, suggesting that LYTIP may be the main source of roasted aroma in TALMT. p-Cymene, which imparted a fruity aroma, was present in TPIP and LYTIP but was considerably higher in TPIP. Linalool, most abundant in TPIP, suggested that the fruity aroma of TALMT (primarily citrus-like) primarily originated from TPIP and LYTIP, with a notable contribution of TPIP (Zhai et al., 2022). Isoborneol and α-terpineol were only detected in TPIP and contributed to woody and herbal aromas, respectively (Sales et al., 2020). Decanal, a fruity aldehyde with a fruity aroma, was present in all ingredients and was the most abundant in LYTIP (Wang et al., 2024). In summary, isoborneol and linalool may impart fruity, woody, and floral aromas to milk tea; decanal may contribute to rich roasted and fruity aromas; 2-methylbutanal and 2-ethyl-6-methylpyrazine may contribute a moderate roasted aroma; and p-cymene and α-terpineol may have significant influences on the herbal and woody aromas, respectively. In addition, the roasted aroma of TALMT is likely to be mainly contributed by LYTIP, the fruity aroma is provided by both TPIP and LYTIP, and TPIP is mainly responsible for the woody and herbal aromas. The rational proportioning of ingredients in OPF enabled the aroma characteristics of TPIP, LYTIP, MP, and SGZCS to be highly integrated, ultimately resulting in a harmonious and pleasant quality of roasted, woody, and fruity aromas. Analysis of taste and color quality of samples Taste profile difference analysis The taste contour analysis was conducted on the three milk tea formulations (PF1, PF2, and OPF) and their ingredients (LYTIP, TPIP, MP, and SGZCS) using an E-tongue system. The results indicated that all three milk tea formulations exhibited a strong umami and astringency profile, along with weaker richness, sourness, and sweetness (Fig. 2C). However, OPF demonstrated a higher taste intensity and a richer taste, particularly in sweetness, which was considerably higher than the other two formulations. Notably, although the astringency and bitterness of OPF were like those of the other two formulations, the sensory score of OPF (85.75) was higher than that of PF1 (85.50) and PF2 (84.65). This difference could be partly attributed to the strong sweet and umami flavor of OPF. Stronger sweetness and umami sensations could mask the bitterness and astringency through taste–taste interactions, thereby creating a more balanced and acceptable overall taste profile (Li et al., 2024). We further observed that the masking effect was related to the antagonistic influence of protein–polysaccharide interactions on bitter taste in the milk tea complex system. Previous studies have shown that these interactions could bind to bitter and astringent compounds via hydrogen bonding and electrostatic interactions, reducing their affinity for taste receptors and effectively suppressing bitterness and astringency (Jauregi et al., 2021; Kan et al., 2023). To investigate the sources of taste differences among the three formulations and to provide a theoretical basis for optimizing milk tea flavor, E-tongue analysis was conducted on the individual ingredients of the milk tea. As shown in Fig. 2D, marked taste differences exist among the ingredients. LYTIP exhibited the strongest aftertaste, richness, astringency, and bitterness; SGZCS contributed the most to sweetness; and TPIP, together with SGZCS, contributed to notable sourness. These results indicated that LYTIP, as the primary raw material of milk tea, plays a major role in contributing to its rich taste. In addition, SGZCS and TPIP played an important role in contributing to the sweetness and sourness of the milk tea. Owing to variations in the amounts of ingredients added across formulations, distinct taste differences were observed. The higher levels of LYTIP and SGZCS in OPF could partly explain its rich and harmonious taste. In summary, OPF exhibited better taste intensity and a richer flavor profile, particularly with respect to sweetness, which was substantially higher than that of the other two formulations. The superior taste quality of OPF was primarily attributed to its enhanced sweetness and umami flavor. The stronger umami and sweetness were able to effectively mask the presence of bitterness and astringency. As the key ingredient in milk tea, LYTIP highly contributed to the overall rich taste, while SGZCS and TPIP primarily contributed to the sweetness and sourness of the milk tea. Qualitative and quantitative analyses of catechins in milk tea As key components of milk tea, catechins play a major role in contributing to its bitterness, astringency, and health benefits (Guo et al., 2021). The differences in catechin content among formulations and ingredients may provide a chemical basis for taste differences in milk tea, especially bitterness and astringency. To clarify their role in determining the quality of TALMT, targeted quantification of catechins was conducted in the three formulations and in LYTIP. The results indicated that the contents of major catechins—EGC, C, EGCG, GCG, and ECG—showed a clear increasing trend with the increase in the addition of LYTIP (Fig. 5). Among the three formulations, the total catechin content was the highest in the OPF, which may suggest potential antioxidant and health-related benefits (Shaukat et al., 2023). Concurrently, the interaction between the sweetness of catechins and SGZCS as well as the mellow taste of TPIP may contribute to the rich taste quality of OPF, supporting its designation as the optimal formulation among the three. The combined effects of catechins, SGZCS, and the mellowing influence of TPIP further may enhance the taste richness of OPF. The structural features of ester catechins led to a lower taste threshold, thereby contributing even better to the bitterness and astringency of tea (Zhang et al., 2020). Non-ester catechins (C and EC) were related to the sweetness and sweet aftertaste of tea, and their high content in the OPF may contribute to the pleasant mouthfeel of the tea (Zhang et al., 2016). Part of the taste performance of OPF arises from complex interactions between ingredients and catechins in the milk tea formulation as well as between ester and non-ester catechins within the formulation matrix.Fig. 5Multivariate statistical analysis based on volatiles. Note: (A) Loading plot of principal component analysis; (B) Hierarchical cluster analysis; (C) Loading plot of partial least squares regression analysis; (D) Permutation test; Two better formulations (PF1, PF2), optimal formulation (OPF).Fig. 5 Multivariate statistical analysis based on volatiles. Note: (A) Loading plot of principal component analysis; (B) Hierarchical cluster analysis; (C) Loading plot of partial least squares regression analysis; (D) Permutation test; Two better formulations (PF1, PF2), optimal formulation (OPF). Notably, the total catechin content in ingredients (LYTIP, 195.73 mg/100 mL) was substantially higher than that in OPF despite identical levels of tea powder input. This reduction in extractable catechins in OPF could be attributed to complexation between catechins and proteins provided by MP. The hydroxyl-rich catechins usually bind to protein molecules via hydrogen bonding and hydrophobic interactions, forming stable complexes that reduce the concentration of freely detectable catechins in milk tea (Zhou et al., 2025). Moreover, such interactions have been reported to enhance the bioavailability of dietary micronutrients by facilitating intestinal absorption, thereby may increase the potential physiological benefits of catechins (Bhagat et al., 2019; Piyasena, 2025). In conclusion, optimizing the ingredient combination and preserving catechins may positively affect the product's taste quality and potential health-related properties. These insights provided a theoretical basis for developing a new milk tea that simultaneously possessed nutritional functionality and high sensory appeal. Analysis of color quality of samples Color is recognized as one of the important factors that influence the sensory quality of milk tea. To understand the relationship between the color values (L*, a*, b*, C*, and H°) of milk tea and its quality, we measured the color values of three TALMT formulations (Table S14). As shown in Table S14, under identical brewing conditions, PF1, PF2, and OPF exhibited significant differences in color values (p < 0.05). All three formulations displayed high L* values and b* values, indicating that they exhibited the bright yellow appearance. However, the three formulations differed in the balance between color values. PF2 had the highest a* and b* values (4.86 ± 0.01 and 30.50 ± 0.01) and the highest C* value (30.88 ± 0.01), indicating that among the three formulations, it exhibited the strongest redness, yellowness, and overall chromaticity. In contrast, OPF exhibited the highest L* value (77.13 ± 0.13), indicating a brighter and lighter appearance, while its a* and b* values (3.72 ± 0.03 and 29.25 ± 0.03) fell between those of PF1 and PF2. Overall, OPF displayed more balanced visual characteristics, with higher brightness and slightly lower chroma than PF2 (Wei et al., 2023). Consumer validation of TALMT This study conducted a questionnaire survey to assess consumer acceptance of milk tea products via collecting 150 valid anonymous responses. Among the respondents, 90.00% reported regular consumption of milk tea, while 10.00% consumed it infrequently. The primary reasons for low consumption frequency were perceived to be excessive sweetness and health concerns (Fig. S4). Regarding product format, 66.67% of respondents preferred freshly brewed cup-style milk tea, while 32.67% favored instant bagged or sachet-style products. The results suggested that the positioning of the product as a low-sugar, health-oriented option aligned with consumer preferences. In the future, adopting a more appealing freshly brewed cup-style format could be considered. This survey provided preliminary insights into refining product positioning and informing subsequent market promotion strategies. To further assess the consumers' most genuine acceptance of the OPF, a five-point Likert scale was employed and visualized using stacked bar charts (Fig. 6) (Martínez-Mendoza et al., 2020). In terms of aroma quality, 45.33% and 40.67% of consumers perceived the aromas of TP and LYT in the OPF as a large contribution, respectively, with a notable influence on the overall aroma. In terms of taste quality, 49.33% and 44.00% of consumers perceived the taste of LYT and TP in the OPF to be the major contributor, with a considerable influence on the overall taste. Moreover, 36.67% of the consumers believed that MP contributed the least to the overall taste of the OPF. In terms of sweetness, 78.66% of respondents rated the product as “slightly sweet” or “moderately sweet,” and 68.67% described it as “slightly sweet” or “not sweet,” reflecting a generally low sweetness perception that aligned with the low-sugar formulation of the product. Notably, 98.67% of consumers expressed a positive attitude toward the product, indicating its high overall acceptability and promising commercial potential. Overall, these results indicate that TALMT exhibited favorable quality with strong consumer acceptance, underscoring its potential competitiveness in the milk tea market (Fig. 7).Fig. 6Targeted detection of catechin content (mg/100 mL) in different formulations and LYTIP. Note: Two better formulations (PF1, PF2); optimal formulation (OPF); large-leaf yellow tea instant powder (LYTIP). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)Fig. 6Fig. 7Five points scale diagram of consumer validation. Note: The contribution and intensity of aroma and taste are 1–5, which represent very small, small, moderate, large, and very large, respectively. The overall sweetness is 1–5, which represent very sweet, sweet, moderately sweet, slightly sweet, not sweet; Overall acceptance is 1–5, which represent very dissatisfied, less satisfied, satisfied, more satisfied, very satisfied.Fig. 7 Targeted detection of catechin content (mg/100 mL) in different formulations and LYTIP. Note: Two better formulations (PF1, PF2); optimal formulation (OPF); large-leaf yellow tea instant powder (LYTIP). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) Five points scale diagram of consumer validation. Note: The contribution and intensity of aroma and taste are 1–5, which represent very small, small, moderate, large, and very large, respectively. The overall sweetness is 1–5, which represent very sweet, sweet, moderately sweet, slightly sweet, not sweet; Overall acceptance is 1–5, which represent very dissatisfied, less satisfied, satisfied, more satisfied, very satisfied. Conclusion This study established an optimal formulation (OPF) via a fuzzy mathematical sensory/response surface model. The E-nose analysis revealed that the OPF exhibited a more balanced aroma profile than other formulations. HS-SPME–GC–MS identified 54 volatiles, and six important volatiles (p-cymene, α-terpineol, 2-methylbutanal, decanal, isoborneol, and linalool) may contribute to woody, roasted, fruity and herbal aromas. The results indicated that different ingredients had different contributions to the aroma of the milk tea. The roasted aroma was likely to be mainly contributed by LYTIP, the fruity aroma was provided by both TPIP and LYTIP, and TPIP was mainly responsible for the woody and herbal aromas. E-tongue results indicated higher overall taste intensity for the OPF, with enhanced sweetness, umami, and taste richness. LYTIP markedly contributed to taste richness, while SGZCS and TPIP imparted sweet and sour taste. The moderate increase in catechin content brought about by LYTIP may have a certain correlation with the improvement of sensory quality. Computer-vision analysis indicated that OPF exhibited balanced color quality. Finally, consumer validation confirmed strong acceptance and high overall quality of OPF. However, this study still lacks validation through methods such as GC-O analysis, matrix-specific odor thresholds, aroma verification experiments, and computer simulations for the exploration of important volatiles. OTs in water provide a practical reference for screening important aroma compounds in milk tea, with the caveat that matrix effects influence their actual sensory contribution. Furthermore, in this study, the focus on non-volatile compounds in milk tea was mainly on the relatively abundant catechins, and there was a lack of non-targeted detection of non-volatile compounds. Future studies are needed to further elucidate the flavor-formation mechanism in detail. Overall, this study comprehensively characterized the formulation optimization, flavor profile, key influencing factors, ingredient contributions, and acceptance of TALMT through an interdisciplinary analytical framework. This study provided theoretical support for the development and research on the flavor of LYT milk tea. CRediT authorship contribution statement Wei Zhao: Writing – original draft, Software, Methodology, Formal analysis, Data curation, Conceptualization. Zhenbin Chen: Software, Methodology, Formal analysis. Yanqun Jiang: Methodology, Investigation, Data curation. Caiyan Sheng: Visualization, Supervision, Investigation. Mingxia Lu: Writing – review & editing, Validation, Supervision. Jixin Zhang: Validation, Supervision. Chunzhao Gong: Supervision, Resources. Chenyang Shao: Writing – review & editing, Validation, Supervision. Jingming Ning: Resources, Project administration, Funding acquisition. Ethical statements The authors confirm that the appropriate protocols for protecting the rights and privacy of all participants were utilized during the execution of this research, including no coercion to participate, full disclosure of study requirements and risks, verbal consent of participants, no release of participant data without their knowledge, and the ability to withdraw from the study at any time. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.