Fresh-cut jack fruit shelf life prediction method based on characteristic metabolites

Abstract

The invention provides a fresh-cut jack fruit shelf life prediction method based on characteristic metabolites, relates to the technical field of food preservation, and solves the problems that the existing shelf life prediction method depends on subjective evaluation or single physicochemical indexes, so that the precision is insufficient, and environmental parameters cannot be dynamically integrated. The method comprises the following steps: detecting metabolite components of fruit bracts with different freshness through liquid nitrogen grinding in combination with an HS-SPME / GC-MS technology, and screening a VIP value gt by using an OPLS-DA model; a BP neural network model is constructed, the concentration of the metabolites and storage parameters are input, the shelf life is predicted, and the model is verified and corrected by comparing a predicted value with an actual value. According to the method, the warehouse logistics strategy can be dynamically optimized, and the quality loss of the fresh-cut jackfruits is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of food preservation. More specifically, the present invention relates to a method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites. Background Art

[0002] Fresh-cut jackfruit is prone to quality deterioration during storage and circulation, and its shelf life is affected by the complex interaction of metabolite changes and environmental factors. In the prior art, shelf life prediction methods mainly rely on sensory evaluation or the monitoring of single physicochemical indices (such as hardness, pH value), and these methods have significant limitations. Sensory evaluation is highly subjective, greatly influenced by the differences in personnel experience, and cannot achieve quantitative prediction. Although physicochemical indices can reflect some quality changes, it is difficult to capture the early signals of dynamic metabolite degradation, resulting in a lag in early warning and unable to effectively guide the real-time adjustment of the supply chain.

[0003] Traditional metabolite detection methods have problems with the loss of volatile components in the sample processing stage. Conventional grinding techniques are mostly carried out at room temperature or ordinary freezing conditions, which cannot completely inhibit enzyme activity and oxidation reactions, resulting in partial degradation or volatilization of volatile metabolites (such as esters, aldehydes) during the preparation process. In the existing methods, it is difficult to balance the sample fragmentation efficiency and metabolite stability: excessive grinding may cause local heating, while insufficient fragmentation leads to incomplete release of cell contents, both of which will affect the representativeness of the detection data. In addition, the temperature fluctuations and loose control of exposure time during the sample transfer process further exacerbate the risk of metabolite profile distortion.

[0004] In terms of differential metabolite screening, existing studies mostly rely on statistical significance thresholds (such as P value) or single-fold change criteria, lacking multi-dimensional stability verification. For example, screening markers only by the fold change between groups may miss metabolites with important predictive value for shelf life but with small change amplitudes; conversely, over-reliance on statistical significance is vulnerable to experimental batch errors or instrument noise interference, resulting in the inclusion of false positive markers. More critically, most methods do not systematically evaluate the repeatability of metabolite detection (such as the coefficient of variation within groups) and the stability of instrument analysis (such as retention time drift), making the screening results have poor reproducibility when verified across platforms or batches, limiting the generalization ability of the prediction model.

[0005] Existing shelf-life prediction models have obvious deficiencies in handling environmental parameters. Most models use static input parameters and cannot integrate the temperature and humidity fluctuation data of the storage environment in real time. For example, the effect of temperature on the metabolite degradation rate is usually simplified to a linear relationship, ignoring the exponential correlation between temperature and reaction rate in the Arrhenius equation, resulting in the predicted value deviating from the actual shelf life when the temperature changes suddenly. In addition, the dual effects of humidity on microbial growth and fruit water loss are not quantified and modeled, making it difficult to accurately reflect the mildew risk in a high-humidity environment or the change of water loss rate under low-humidity conditions. This static modeling method makes the prediction results unable to adapt to the dynamic environmental changes during storage and transportation, reducing the timeliness of supply chain management strategies.

[0006] Insufficient biological replicates is another technical difficulty. Some studies did not set up enough biological replicate samples or did not strictly screen the consistency of data within the group, resulting in the metabolite concentration data being greatly affected by individual differences, sampling position deviation or operation errors. For example, factors such as the maturity gradient of fresh-cut jackfruit bracts and local microbial colonization differences may introduce non-systematic variations. If not controlled by sufficient replicate samples and coefficient of variation screening, it will directly affect the reliability of differential metabolite screening and further reduce the accuracy of the prediction model.

[0007] The root cause of the above problems lies in the lack of a systematic solution in the existing technology: it is necessary to achieve high-fidelity extraction of metabolites in the sample preparation process, establish multi-dimensional screening criteria to improve the stability of markers, and require the prediction model to have the ability to dynamically integrate environmental parameters. However, there are multiple technical bottlenecks in achieving these goals. For example, the equipment cost and operation complexity of maintaining a low-temperature liquid nitrogen environment restrict large-scale applications; the heterogeneity of multi-dimensional data (metabolites, environment, quality indicators) increases the difficulty of model training; the spatial resolution and real-time data transmission of environmental sensors are insufficient, making it difficult to construct a high-precision three-dimensional parameter grid. These challenges have slowed down the practical process of the shelf-life prediction technology for fresh-cut jackfruit, and there is an urgent need for a breakthrough at the methodological level. Summary of the Invention

[0008] An object of the present invention is to provide a method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites, so as to optimize the supply chain management strategy, ensure that the fresh-cut jackfruit maintains good quality during the shelf life, and reduce losses.

[0009] To achieve these and other advantages of the present invention, according to one aspect of the present invention, the present invention provides a method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites, including the following steps: Fresh-cut jackfruit bract samples with different freshness levels were obtained. Each sample was ground in liquid nitrogen and placed in a headspace vial. Saturated NaCl solution and internal standard solution were added. The metabolites in the sample were detected by using the fully automatic headspace solid-phase microextraction (HS-SPME) combined with gas chromatography-mass spectrometry (GC-MS) technology. Based on the orthogonal partial least squares discriminant analysis (OPLS-DA) model, among the metabolite components, the differential metabolites of fresh-cut jackfruit under different freshness levels were screened. The screening criteria were as follows: a) The variable importance in the projection (VIP) value > 1; b) The fold change of the metabolite in fresh-cut jackfruit under different freshness levels ≥ 2 or ≤ 0.5; The metabolite components that simultaneously met conditions a) and b) were screened as differential metabolites; Using the screened differential metabolites as markers, a backpropagation (BP) neural network prediction model was constructed. The concentration data of the differential metabolites and the storage environment parameters were input, and the predicted value of the shelf life of fresh-cut jackfruit was output. By comparing the predicted result data of the prediction model with the actual shelf life data, the accuracy of the marker screening was verified, and the markers were further screened and the prediction model was corrected accordingly.

[0010] Through the combination of grinding in liquid nitrogen and HS-SPME / GC-MS technology, the present invention effectively preserves the integrity of volatile metabolites in fresh-cut jackfruit and avoids the interference of high-temperature degradation on the detection results. By using the OPLS-DA model to screen differential metabolites with VIP value > 1 and significant fold changes, the key markers related to the shelf life can be accurately located. By constructing a BP neural network model to integrate multi-source data, the dynamic prediction of the shelf life can be realized, providing a scientific basis for supply chain management and reducing the quality fluctuations and losses caused by subjective judgment.

[0011] Preferably, after constructing a backpropagation (BP) neural network prediction model using the screened differential metabolites, the model is trained and optimized. After the training is completed, the metabolite data of fresh-cut jackfruit collected in real time are input into the model, and combined with the storage temperature and humidity parameters monitored in real time, the predicted result of the shelf life is dynamically adjusted. According to the dynamically adjusted predicted result of the shelf life, the supply chain management strategy is optimized, including but not limited to adjusting the logistics transportation time, storage conditions and sales rhythm, so as to ensure that fresh-cut jackfruit maintains good quality during the shelf life and reduce losses.

[0012] In the present invention, the dynamic adjustment mechanism automatically corrects the predicted shelf life value with environmental fluctuations by collecting temperature and humidity data in real time and model feedback, improving the prediction accuracy. The weight matrix is updated regularly in combination with the incremental learning algorithm to enhance the adaptability of the model to seasonal or regional environmental changes. The optimized supply chain strategy can specifically adjust the warehousing and logistics plans to ensure that fresh-cut jackfruit maintains the best quality state during transportation and sales.

[0013] Preferably, the GC-MS detection conditions include: Chromatographic conditions: DB-5MS capillary column 30 m × 0.25 mm × 0.25 μm, the carrier gas is high-purity helium, the constant flow rate is 1.2 mL / min, the inlet temperature is 250 °C, splitless injection, solvent delay 3.5 min. Programmed temperature rise: maintain at 40 °C for 3.5 min, rise to 100 °C at 10 °C / min, then rise to 180 °C at 7 °C / min, and finally rise to 280 °C at 25 °C / min, hold for 5 min; Mass spectrometry conditions: electron impact ionization source EI, ion source temperature 230 °C, quadrupole temperature 150 °C, mass spectrometry interface temperature 280 °C, electron energy 70 eV, the scanning mode is selected ion monitoring mode SIM, and qualitative and quantitative ions are accurately scanned.

[0014] In the present invention, the combination of the DB-5MS chromatographic column and the EI ion source optimizes the metabolite separation efficiency and ensures the accuracy of qualitative and quantitative analysis of volatile components. The programmed temperature rise gradient design (rising from 40 °C to 280 °C) takes into account the detection requirements of low-boiling and high-boiling metabolites and reduces peak overlap interference. The SIM scanning mode reduces the influence of background noise on the detection results by accurately selecting characteristic ions and improves the data reliability.

[0015] Preferably, the screening of differential metabolites further includes: The screened common differential metabolites are plotted into a metabolite abundance box plot to show the changing trends of differential metabolites under different freshness levels.

[0016] In the present invention, the drawing of the metabolite abundance box plot visually shows the changing trends of markers at different fresh stages, facilitating the rapid identification of key quality indicators. Through visual analysis, the effectiveness of the screening criteria can be verified, assisting researchers in optimizing the metabolite screening threshold and further improving the robustness of model prediction.

[0017] Preferably, the input parameters of the BP neural network model include: The concentration data of differential metabolites determined by the OPLS-DA model and multiple screening criteria, including the concentration data of volatile and non-volatile differential metabolites with significant differences under different freshness levels; The storage temperature parameters and storage time cumulative parameters monitored in real time, where the temperature parameters are accurate to 0.1 °C and the time parameters are in hours, recording the storage temperature and time parameters; Include the core indicators characterizing fruit quality, including soluble solids content, measured by a refractometer with an accuracy of ±0.5° Brix; total acid content, calculated as tartaric acid, determined by acid-base titration; and pH value, with an accuracy of ±0.01.

[0018] In the present invention, the input parameters cover differential metabolite concentrations, storage environment parameters, and core quality indicators, improving the comprehensiveness of model prediction through multi-dimensional data coupling. The temperature parameter accuracy of ±0.1 °C and the refined recording of time cumulative parameters strengthen the quantitative correlation between environmental factors and the metabolite degradation rate, avoiding prediction deviations caused by a single variable.

[0019] Preferably, when screening differential metabolites based on the OPLS-DA model, the number of biological replicates is set to ≥ 3; In univariate analysis, the P value or FDR value is used as the statistical significance determination index, where the P value needs to satisfy P < 0.05, or the FDR value is controlled at ≤ 0.05, and the number of biological replicates corresponding to univariate analysis is ≥ 2.

[0020] In the present invention, setting the number of biological replicates to ≥ 3 and combining P value / FDR screening significantly improves the statistical credibility of differential metabolite screening. The multiple verification mechanism (P < 0.05 or FDR ≤ 0.05) in univariate analysis effectively excludes random error interference, ensuring the relevance of the screening results to the biological significance and actual shelf life.

[0021] Preferably, the screening step with variable importance in projection VIP value > 1 includes: Based on ≥ 3 groups of independent biological replicate samples, each group containing more than 5 fresh-cut fruit bracts, calculate the coefficient of variation CV of the metabolite peak areas within the group, and only retain metabolites with CV ≤ 15% for OPLS-DA modeling; For volatile metabolites detected by GC-MS, calculate the retention time RT drift through quality control samples. For metabolites with a retention time RT deviation ≤ 0.15 min and a relative standard deviation RSD < 2%, multiply the variable importance in projection VIP value by a stability coefficient of 1.2; For metabolites with a fold change ≥ 2 or ≤ 0.5 in fresh-cut jackfruit at different freshness levels, add a differential enhancement factor of 0.3 to the variable importance in projection VIP value; Calculate the standard deviation SD of the variable importance in projection VIP values for 3 independent experiments, and only retain metabolites with a standard deviation SD < 0.1 for the variable importance in projection VIP value, excluding candidate markers with poor repeatability; For metabolites with variable importance projection VIP > 1, the peak shape symmetry factor As is required to be between 0.9 and 1.1, and the signal-to-noise ratio S / N > 30, excluding the interference of instrument noise; Introduce 200 permutation tests to ensure the predictive ability Q of metabolites with variable importance projection VIP > 1 in the permutation model 2 The intercept < 0.05 to prevent false positives caused by overfitting.

[0022] In the present invention, through multi-level quality control such as CV ≤ 15%, RT drift correction, and signal-to-noise ratio screening, the influence of experimental fluctuations and instrument noise on the VIP value is greatly reduced. The application of the differential enhancement factor and permutation test enhances the ability to identify significantly different metabolites, while suppressing the risk of model overfitting, ensuring the robustness and repeatability of biomarker screening.

[0023] Preferably, the process of placing each sample in a headspace vial after grinding with liquid nitrogen and adding saturated NaCl solution and internal standard solution specifically includes: Use a cryogenic grinder with a frequency of 18 - 20 Hz to grind fresh cut bracts with liquid nitrogen. The single grinding time is controlled within 30 - 45 s. During the grinding process, continuously supplement liquid nitrogen to maintain the temperature ≤ -196 °C, ensuring that the cell breakage rate ≥ 95%, and at the same time avoiding the degradation of volatile metabolites caused by high temperature; Immediately transfer the ground sample to a pre-cooled 50 mL headspace vial at -80 °C, and the transfer time < 15 s; Add the solution according to the ratio of sample mass g: saturated NaCl solution volume mL = 1:4. Utilize the salting-out effect with an ionic strength ≥ 6 mol / L to promote the transfer of volatile metabolites from the aqueous phase to the headspace phase; Perform ultrasonic pretreatment at 40 kHz with a power of 200 W for 5 min; Add 20 μL of deuterated internal standard solution with a concentration of 10 μg / mL, and the retention time difference from the target metabolite ≥ 2 min; Add 0.1 % v / v ascorbic acid to the internal standard solution.

[0024] In the present invention, liquid nitrogen grinding (temperature ≤ -196 °C) combined with the salting-out effect (ionic strength ≥ 6 mol / L) maximizes the extraction of volatile metabolites and inhibits oxidative degradation. Ultrasonic pretreatment (40 kHz) promotes the release of metabolites, and the addition of deuterated internal standard and ascorbic acid effectively corrects instrument drift, ensuring the accuracy of detection data and batch-to-batch consistency.

[0025] Preferably, the HS-SPME detection conditions include: Extraction was carried out using a SPME Arrow extraction head, and the extraction conditions were controlled as follows: constant temperature oscillation at 60 °C for 5 min, oscillation rate of 200 rpm; during headspace extraction for 15 min, the extraction head was kept 5 mm away from the liquid surface of the sample; desorption at 250 °C for 5 min.

[0026] In the present invention, the use of the SPME Arrow extraction head (120 μm coating) improves the enrichment efficiency of volatile metabolites by optimizing the adsorption phase and extraction time (15 min). The precise control of constant temperature oscillation (60 °C) and desorption temperature (250 °C) balances the extraction rate and the risk of thermal decomposition, ensuring detection sensitivity and stability.

[0027] Preferably, the method for dynamically adjusting the shelf-life prediction results in combination with the monitored storage temperature and humidity parameters includes: Deploy a temperature and humidity sensor array in the storage environment with a temperature accuracy of ± 0.1 °C and a humidity accuracy of ± 2 % RH. Collect data in real time at intervals of every 5 min and synchronize it to the data processing terminal through the LoRa wireless transmission module to construct a three-dimensional environmental parameter grid covering the storage space with a spatial resolution of ≤ 0.5 m 3 ; Based on the Arrhenius equation, fit the degradation rate of characteristic metabolites at different temperatures. When the real-time temperature deviates from the preset storage temperature by ± 2 °C, automatically adjust the metabolite concentration decay coefficient in the BP neural network with an adjustment step size of ≤ 0.05 min; When the humidity > 85% RH, trigger a mold growth warning and superimpose a shelf-life reduction coefficient of 0.8 - 0.9 on the output layer of the model; when the humidity < 60% RH, increase the water loss rate compensation parameter by 10% - 15% to correct the quality prediction deviation caused by water loss; Adopt a sliding window incremental learning algorithm. Every 24 h, input the latest 50 groups of temperature and humidity data and the corresponding concentration values of differential metabolites into the BP neural network to locally update the weight matrix related to environmental parameters, ensuring the model's adaptive ability to seasonal environmental fluctuations and dynamically converging the prediction error to ≤ 5%.

[0028] In the present invention, the three-dimensional temperature and humidity grid (resolution ≤ 0.5 m 3 ) combined with the Arrhenius equation realizes the spatial fine modeling of the metabolite degradation rate. The humidity-triggered reduction coefficient and compensation parameter dynamically correct the prediction deviation, and the incremental learning algorithm (updating 50 groups of data every 24 h) ensures that the model continuously adapts to environmental changes and maintains the convergence state of the prediction error ≤ 5%. Brief Description of the Drawings

[0029] Figure 1Flow chart of an embodiment of the present application; Figure 2 Schematic diagram of a fresh cut fruit bract in a rigid state according to an embodiment of the present application; Figure 3 Schematic diagram of a fresh cut fruit bract in an immature state according to an embodiment of the present application; Figure 4 Schematic diagram of a fresh cut fruit bract in a mature state according to an embodiment of the present application; Figure 5 Schematic diagram of a fresh cut fruit bract in an over - mature state according to an embodiment of the present application; Figure 6 Box plot of the differential metabolite (E)-2 - methylbutyric acid 3,7 - dimethyl - 2,6 - octadienyl ester according to an embodiment of the present application; Figure 7 Box plot of the differential metabolite 1H - pyrrole - 2 - carbonitrile according to an embodiment of the present application; Figure 8 Box plot of the differential metabolite 5 - ethyl - 2 - heptanol according to an embodiment of the present application; Figure 9 Box plot of the differential metabolite isopentyl acetoacetate according to an embodiment of the present application; Figure 10 Box plot of the differential metabolite phenethyl 3 - methylbutyrate according to an embodiment of the present application; Figure 11 Box plot of the differential metabolite ethyl isonicotinate according to an embodiment of the present application; Figure 12 Box plot of the differential metabolite quinoxaline according to an embodiment of the present application; Figure 13 Box plot of the differential metabolite 4 - methyl - 1 - (1 - methylethyl)bicyclo(3.1.0) - 3 - hexen - 2 - one according to an embodiment of the present application; Figure 14 Box plot of the differential metabolite 5 - butyldihydro - 2(3H) - furanone according to an embodiment of the present application; Figure 15 Box plot of the differential metabolite (Z)-1 - (1 - methoxyethoxy) - 3 - hexene according to an embodiment of the present application; Figure 16 Box plot of the differential metabolite 1H - pyrrole - 3 - carbonitrile according to an embodiment of the present application; Figure 17 Box plot of the differential metabolite phenethyl 2 - methylbutyrate according to an embodiment of the present application. Detailed implementation manners

[0030] The following further elaborates on the present invention in conjunction with the accompanying drawings, enabling those skilled in the art to implement it based on the text of the specification.

[0031] As Figure 1 shown, the present invention provides a method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites, including the following steps: S1. Obtain fresh-cut jackfruit bract samples of different freshness levels. After each sample is ground with liquid nitrogen and placed in a headspace vial, saturated NaCl solution and internal standard solution are added. The metabolite components in the samples are detected by using fully automatic headspace solid-phase microextraction (HS-SPME) combined with gas chromatography-mass spectrometry (GC-MS) technology. Specifically, the process of obtaining fresh-cut jackfruit bract samples of different freshness levels includes: Pick the eight-ripe fresh "Thai No. 8" jackfruit from Sandao Town, Baoting Li and Miao Autonomous County, Hainan Province, China, and transport it to the National Southern Crop Improvement Institute, Chinese Academy of Agricultural Sciences, Yazhou District, Sanya City, Hainan Province. The fruits should be selected according to the criteria of consistent color, uniform size, no mechanical damage on the surface, and no pest or disease infestation. After selection and grouping, subsequent experiments are carried out. Place them at the simulated room temperature (22 °C), and on the 0th day, 3rd day, 6th day, and 9th day respectively, and under the condition of a relatively clean space (26 °C), cut the jackfruit along the main axis with a sterilized stainless steel knife. After removing the fruit stalk, take out the complete bract, weigh it separately and put it into a covered fresh-keeping box. Put the loaded jackfruit in storage at different temperatures (10 °C, normal temperature), and randomly select the measurement indexes and take samples at 0, 1, and every 1 day for subsequent experiments.

[0032] Select the following four different state points through the physiological and biochemical indexes of the bracts to measure their volatile substance components: 1. Lignification state: Since the maturity of the jackfruit fruit is relatively low when the whole fruit is stored for 0 days, and the bract does not rot but the surface begins to turn white on the 7th day when stored at low temperature. Select the bract of the whole fruit stored for 0 days and on the 9th day of the low-temperature shelf life as the sample point (Lignification) of the lignification phenomenon, as Figure 2 shown.

[0033] 2. Unripe: The fresh-cut jackfruit bracts stored at low temperature with the whole fruit stored for 3 days can maintain the change of bract hardness and inhibit the rising rate of conductivity and total phenol content. Fresh-cut jackfruit with a soluble solid content of less than 11% and a total acid content of less than 20 g / kg is in an unripe state and has not reached the suitable range for eating. Therefore, select the state (Unripe) that is still unripe on the 1st day of the low-temperature shelf life with the whole fruit stored for 3 days, as Figure 3 shown.

[0034] 3. Perfectly-ripe: The hardness of fresh-cut jackfruit in a perfectly-ripe state is approximately 10 N - 12 N, the soluble solids content is above 15%, and the total acid content is 23 g / kg. The surface color of the fresh-cut fruit bracts of the whole fruit stored for 6 days and with a low-temperature shelf life of 3 days is bright and vivid, the smell is more intense, the soluble solids content is at the highest point, and the total acid is in a steadily rising stage. Therefore, this point is selected as the sample point of perfectly-ripe (Perfectly-ripe), as Figure 4 shown.

[0035] 4. Decay: The jackfruit of the whole fruit stored for 9 days has too high a degree of ripeness of the whole fruit. Although its ripe aroma is very intense and the fruit is sweet, its shelf life is too short. When the whole fruit is stored for 9 days and the low-temperature shelf life is 4 days, the fruit bracts are intact but become soft, and the total acid content and soluble solids content are relatively high. Therefore, this point is selected as the sample point of over-ripe state (Decay), as Figure 5 shown.

[0036] Specifically, the process of placing each sample in a headspace vial after liquid nitrogen grinding and adding saturated NaCl solution and internal standard solution includes: Using a cryogenic grinder with a frequency of 18 Hz, perform liquid nitrogen grinding on the fresh-cut fruit bracts, control the single grinding time within 40 s, continuously supplement liquid nitrogen during the grinding process to maintain the temperature ≤ -196 °C, ensure that the cell breakage rate ≥ 95%, and at the same time avoid the degradation of volatile metabolites caused by high temperature; Weigh approximately 500 mg of the ground sample and immediately transfer it to a pre-cooled 50 mL headspace vial at -80 °C, and the transfer time < 15 s; Add the solution according to the ratio of sample mass g: saturated NaCl solution volume mL = 1:4. Utilize the salting-out effect, with an ionic strength ≥ 6 mol / L, to promote the transfer of volatile metabolites from the aqueous phase to the headspace phase; Perform ultrasonic pretreatment at 40 kHz and a power of 200 W for 5 min; Add 20 μL of deuterated internal standard solution with a concentration of 10 μg / mL (ethyl acetate-d4, or methyl benzoate-d3 can also be selected), and the retention time difference from the target metabolite ≥ 2 min; Add 0.1% v / v ascorbic acid to the internal standard solution.

[0037] Specifically, the detection conditions of the fully automatic headspace solid-phase microextraction HS-SPME include: Extraction was carried out using a SPME Arrow extraction head (120 µm, DVB / CWR / PDMS). The extraction conditions were controlled as follows: constant temperature oscillation at 60 °C for 5 min, oscillation rate of 200 rpm; during headspace extraction for 15 min, the extraction head was kept 5 mm from the sample liquid surface; desorption at 250 °C for 5 min; Before sampling, the extraction head was aged at 250 °C in the Fiber Conditioning Station for 5 min. Note: New extraction heads should be aged in the Fiber Conditioning Station for 2 h before extraction.

[0038] Specifically, the detection conditions for gas chromatography - mass spectrometry (GC - MS) include: Chromatographic conditions: DB - 5MS capillary column, 30 m × 0.25 mm × 0.25 μm, carrier gas is high - purity helium, constant flow rate of 1.2 mL / min, inlet temperature 250 °C, splitless injection, solvent delay 3.5 min. Programmed temperature rise: hold at 40 °C for 3.5 min, rise to 100 °C at 10 °C / min, then rise to 180 °C at 7 °C / min, and finally rise to 280 °C at 25 °C / min and hold for 5 min; Mass spectrometry conditions: electron impact ionization source (EI), ion source temperature 230 °C, quadrupole temperature 150 °C, mass spectrometry interface temperature 280 °C, electron energy 70 eV, scanning mode is selected ion monitoring (SIM), qualitative and quantitative ions are accurately scanned.

[0039] S2. Based on the orthogonal partial least squares discriminant analysis (OPLS - DA) model, among the metabolite components, differential metabolites of fresh - cut jackfruit under different freshness levels were screened. The screening criteria were: a) Variable Importance in Projection (VIP) value > 1; b) Fold - change of metabolites in fresh - cut jackfruit under different freshness levels ≥ 2 or ≤ 0.5; Metabolite components that simultaneously meet conditions a) and b) were screened as differential metabolites; Specifically, first, the screening of differential metabolites for pairwise comparison was carried out. Based on the Variable Importance in Projection (VIP) obtained from the OPLS - DA model (biological replicates ≥ 3), differential metabolites between samples of different freshness levels could be preliminarily screened.

[0040] Meanwhile, the P - value / FDR (biological replicates ≥ 2) or FC value of univariate analysis could be combined to further screen out differential metabolites. The screening criteria for differential metabolites in this application are: 1. Select metabolites with VIP > 1. The VIP value represents the influence intensity of the between-group differences of the corresponding metabolites in the classification and discrimination of each group of samples in the model. Generally, metabolites with VIP > 1 are considered to have significant differences.

[0041] 2. Select metabolites with fold change ≥ 2 and fold change ≤ 0.5. That is, if the difference in metabolites between the control group and the experimental group is more than 2-fold or less than 0.5-fold, it is considered to have a significant difference.

[0042] Using the screening results of pairwise comparisons, re-screen the differential metabolites that exist in each comparison, and make box plots to show the change trends of differential metabolites under different freshness levels. The box plots of the obtained differential metabolites are as Figures 6 - 17 shown: The differential metabolites include: (E)-2-methylbutyric acid 3,7-dimethyl-2,6-octadienyl ester, 1H-pyrrole-2-carbonitrile, 5-ethyl-2-heptanol, isopentyl acetoacetate, 2-phenylethyl 3-methylbutyrate, ethyl isonicotinate, quinoxaline, 4-methyl-1-(1-methylethyl)bicyclo(3.1.0)-3-hexen-2-one, 5-butyldihydro-2(3H)-furanone, (Z)-1-(1-methoxyethoxy)-3-hexene, 1H-pyrrole-3-carbonitrile, 2-phenylethyl 2-methylbutyrate.

[0043] S3. Use the screened differential metabolites as markers to construct a backpropagation BP neural network prediction model. Input the concentration data of the differential metabolites and the storage environment parameters, and output the predicted shelf life value of fresh-cut jackfruit. By comparing the predicted result data of the prediction model with the actual shelf life data, verify the accuracy of marker screening, further screen the markers based on this, and correct the prediction model.

[0044] Specifically, the input parameters of the BP neural network model include: The concentration data of differential metabolites determined by the OPLS-DA model and multiple screening criteria, including the concentration data of volatile and non-volatile metabolites with significant differences under different freshness levels; The storage temperature parameters and storage time accumulation parameters monitored in real time, where the temperature parameter is accurate to 0.1 °C, and the time parameter is in hours, recording the storage temperature and time parameters. For example: Temperature sensors are arranged at the top, middle, and bottom of the storage environment to form a three-dimensional monitoring grid, with a spatial resolution ≤ 0.5 m 3 . The data recording module is connected to the sensor through a shielded cable and fixed on the outer wall of the storage box. The pH meter probe is directly inserted into the bract sample to avoid contact with air. During the working process, the temperature data is transmitted to the data processing terminal through the LoRa module.

[0045] Core indicators representing fruit quality are incorporated, including soluble solids content, measured by a refractometer with an accuracy of ±0.5 °Brix; total acid content, calculated as tartaric acid, determined by acid-base titration; and pH value, with an accuracy of ±0.01.

[0046] Specifically, the model training uses the backpropagation algorithm, with 3 hidden layers, the number of nodes being 16, 8, and 4 respectively, and the activation function being ReLU. The training dataset contains 200 groups of historical data, the validation set is 50 groups, the learning rate is set to 0.001, and the number of iterations is 500 times.

[0047] In terms of equipment selection, the model training can run based on the TensorFlow or PyTorch framework, deployed on an embedded industrial computer with a CPU main frequency ≥ 2.4 GHz and a memory ≥ 8 GB. The humidity sensor uses a capacitive sensor with a range of 0 - 100%RH and an accuracy of ±2%. In terms of materials, the training data is stored using an SSD hard drive with anti-vibration design to adapt to the storage environment. In terms of the assembly location, the industrial computer is installed in the control cabinet of the storage room and connected to the sensor array through Ethernet. The humidity sensors are arranged at the four corners of the area where the fruit bracts are stored.

[0048] During the working process, every 24 hours, the sliding window incremental learning algorithm is adopted, and the latest 50 groups of data are input into the model to update the weight matrix related to the environmental parameters. After the training is completed, the model outputs the predicted shelf life value and is verified by the mean square error (MSE ≤ 0.1) with the actual shelf life data. If the error exceeds the threshold, the differential metabolites are re-screened and the model structure is optimized.

[0049] In the verification stage, the predicted results are compared with the actual shelf life data, and the determination coefficient R 2 ≥ 0.9. The correction methods include excluding metabolites with a VIP value standard deviation > 0.1 and increasing the biological replicates to 5 groups. The sample size for a single experiment is 30 fruit bracts, divided into 3 groups with 10 in each group. The statistical method uses a t-test ( P <0.05) or FDR correction (Q < 0.05).

[0050] In terms of equipment selection, the data comparison can be completed based on the MATLAB or Python's SciPy library and run on the same industrial computer. The statistical analysis uses analysis of variance (ANOVA) with a significance level of α = 0.05. In terms of materials, the experimental samples are "Thai No. 8" jackfruit bracts, picked from Baoting County, Hainan Province, and stored in an environment of 10 °C and room temperature. In terms of the assembly location, the verification data is stored in the local database of the industrial computer, with an independent partition from the data output by the prediction model.

[0051] During the operation, the actual shelf life data is obtained by combining manual sensory evaluation and instrumental testing, including fruit firmness (measured using a texture analyzer), color change (measured using a colorimeter), and mold rate. If the prediction error continuously exceeds 5%, the model correction process is triggered: resampling the samples, optimizing the OPLS-DA screening conditions, and updating the number of input layer nodes of the BP neural network.

[0052] Furthermore, after constructing a backpropagation (BP) neural network prediction model using the selected differential metabolites, the model is trained and optimized. After training, the metabolite data of fresh-cut jackfruit collected in real time is input into the model, and combined with the real-time monitored storage temperature and humidity parameters, the shelf life prediction result is dynamically adjusted. According to the dynamically adjusted shelf life prediction result, the supply chain management strategy is optimized, including but not limited to adjusting the logistics transportation time, storage conditions, and sales rhythm, to ensure that the fresh-cut jackfruit maintains good quality during the shelf life and reduces losses.

[0053] Specifically, the method for dynamically adjusting the shelf life prediction result by combining the real-time monitored storage temperature and humidity parameters includes: An array of temperature and humidity sensors is deployed in the storage environment, with a temperature accuracy of ± 0.1 °C and a humidity accuracy of ± 2% RH. Data is collected in real time at 5-minute intervals and synchronized to the data processing terminal through a LoRa wireless transmission module to construct a three-dimensional environmental parameter grid covering the storage space, with a spatial resolution of ≤ 0.5 m 3 . In terms of equipment selection, the temperature sensor can use a thermocouple or a platinum resistance sensor, the humidity sensor can choose a capacitive or resistive sensor, and the wireless transmission module can support LoRa or NB-IoT communication protocols. In terms of materials, the sensor housing can be made of ABS engineering plastic, and the sealing ring can be made of silicone rubber. In terms of the installation position, the temperature sensor can be installed on the top, middle, and bottom shelf beams of the storage room, and the humidity sensor can be deployed at the four corners and the center of the fruit bract storage area. The LoRa wireless transmission module can be fixed in the storage room control cabinet and connected to the sensor through a shielded cable. During operation, the sensor collects data every 5 minutes, uploads it through the wireless module after filtering, and constructs a real-time updated three-dimensional temperature and humidity distribution map. In the parameter setting method, the coordinate mapping of the three-dimensional grid can be realized based on UWB positioning technology. CRC-16 algorithm is used for data verification, and the packet loss rate is ≤ 0.1%. The sensor calibration period can be set to once every 30 days, and it is calibrated using a standard temperature and humidity source. During experimental verification, an array of sensors is deployed in a constant temperature storage room at 10 °C, and data is continuously collected for 72 hours. The calculated spatial temperature uniformity error is ≤ 0.3 °C; Based on the Arrhenius equation, the degradation rates of characteristic metabolites at different temperatures are fitted. When the real-time temperature deviates from the preset storage temperature by ±2 °C, the attenuation coefficient of metabolite concentration in the BP neural network is automatically adjusted, and the adjustment step size is ≤ 0.05 / min. When the real-time temperature deviates from the preset value by ±2 °C, the adjustment step size of the metabolite concentration attenuation coefficient can be set to ≤ 0.05 / min. The activation energy parameter of the Arrhenius equation can be set to 35 - 45 kJ / mol, and the frequency factor can be taken as 1×10 12 - 5×10 13 min -1 。In terms of equipment selection, the data processing terminal can be equipped with an embedded industrial computer, the CPU main frequency can be ≥ 2.4 GHz, and the memory can be ≥ 8 GB. In terms of materials, the operation data can be stored in an industrial-grade SSD hard disk, and the heat sink can be made of aluminum alloy. In terms of the installation location, the industrial computer can be installed in the storage cabinet and connected to the wireless transmission module through Ethernet. The dynamic parameter adjustment module can be integrated into the algorithm script written in Python or MATLAB. During the working process, the temperature data is input into the Arrhenius equation to calculate the real-time degradation rate, and the output result is transmitted to the BP neural network through the API interface to update the weight parameters of the hidden layer. The experimental method includes testing the metabolite degradation curve at three temperature gradients of 15 °C, 25 °C, and 35 °C, and the determination coefficient R 2 ≥ 0.95. When the parameters are corrected, if the temperature fluctuation lasts for more than 10 minutes, the weight update of the entire model is triggered. The statistical verification uses linear regression analysis, and the sum of squared residuals is ≤ 0.08; When the humidity > 85% RH, a mold growth warning is triggered, and a shelf-life reduction coefficient of 0.8 - 0.9 is superimposed on the output layer of the model; when the humidity < 60% RH, a water loss rate compensation parameter of 10% - 15% is increased to correct the quality prediction deviation caused by water loss. In terms of equipment selection, the warning module can be implemented based on a buzzer or an LED indicator light, and the compensation algorithm can be embedded in the TensorFlow Lite framework. In terms of materials, the humidity probe can be made of polyimide substrate, and the moisture-proof encapsulation uses an epoxy resin coating. In terms of the installation location, the warning indicator light can be installed on the control panel at the entrance of the storage room, and the buzzer can be integrated into the housing of the temperature and humidity sensor. During the working process, when the humidity exceeds the threshold, the reduction coefficient is automatically superimposed on the output layer of the model, and a warning signal is sent to the logistics management system through the CAN bus. After the compensation parameter is updated, the shelf-life prediction value is recalculated, and the error convergence time is ≤ 1 hour. The function test includes simulating the environments of 85% RH and 55% RH in a humidity chamber to verify that the response delay of the reduction coefficient and the compensation parameter is ≤ 2 minutes. In the structural design, the compensation algorithm is implemented using the look-up table method, and the query interval is 1 minute.

[0054] Using a sliding window incremental learning algorithm, every 24 h, the latest 50 groups of temperature and humidity data and the corresponding concentration values of differential metabolites are input into the BP neural network to locally update the weight matrix related to environmental parameters, ensuring the model's adaptive ability to seasonal environmental fluctuations, and the prediction error dynamically converges to ≤ 5%.

[0055] Through the above-mentioned embodiments, through high-precision environmental monitoring, dynamic parameter adjustment, and humidity compensation mechanism, the shelf-life prediction model can be optimized in real time. The system adapts to the temperature and humidity fluctuations in the storage environment, reduces the prediction error caused by environmental deviation, improves the response speed of supply chain management, effectively maintains the quality stability of fresh-cut jackfruit, and reduces the loss risk.

[0056] Furthermore, the screening step with variable importance projection VIP value > 1 includes: Based on ≥ 3 groups of independent biological replicate samples, with each group containing more than 5 fresh-cut fruit bracts, calculate the coefficient of variation CV of the metabolite peak areas within the group, and only retain the metabolites with CV ≤ 15% for OPLS-DA modeling; For the volatile metabolites detected by GC-MS, calculate the retention time RT drift through quality control samples. For the metabolites with a retention time RT deviation ≤ 0.15 min and a relative standard deviation RSD < 2%, multiply the variable importance projection VIP value by a stability coefficient of 1.2; For the metabolites with a fold change ≥ 2 or ≤ 0.5 in fresh-cut jackfruit under different freshness levels, add a differential enhancement factor of 0.3 to the variable importance projection VIP value; Calculate the standard deviation SD of the variable importance projection VIP values of 3 independent experiments, and only retain the variable importance projection VIP values of the metabolites with a standard deviation SD < 0.1, and eliminate the candidate markers with poor repeatability; For the metabolites with variable importance projection VIP > 1, require the peak shape symmetry factor As to be between 0.9 - 1.1 and the signal-to-noise ratio S / N > 30 to exclude instrument noise interference; Introduce 200 permutation tests to ensure that the prediction ability Q of the metabolites with variable importance projection VIP > 1 in the permutation model 2 The intercept < 0.05 to prevent false positives caused by overfitting.

[0057] Through the above-mentioned embodiments, through multi-level stability screening, dynamic correction, and strict statistical verification, the accuracy of differential metabolite screening can be improved. The system effectively excludes instrument noise and experimental fluctuation interference, reduces the misselection of false positive markers, ensures the reliability and generalization ability of the prediction model, and provides a stable data basis for the shelf-life prediction of fresh-cut jackfruit.

[0058] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and embodiments. It can be fully applied to various fields suitable for the present invention. For those skilled in the art, additional modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and the examples shown and described herein.

Claims

1. A method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites, characterized in that, It includes the following steps: Obtain fresh-cut jackfruit bract samples with different freshness levels. After each sample is ground with liquid nitrogen and placed in a headspace vial, saturated NaCl solution and deuterated internal standard solution are added. The metabolite components in the sample are detected by using fully automatic headspace solid-phase microextraction (HS-SPME) combined with gas chromatography-mass spectrometry (GC-MS) technology; Based on the orthogonal partial least squares discriminant analysis (OPLS-DA) model, among the metabolite components, screen the differential metabolites of fresh-cut jackfruit under different freshness levels. The screening criteria are: a) The variable importance in the projection (VIP) value > 1; b) The fold change of the metabolite in fresh-cut jackfruit under different freshness levels ≥ 2 or ≤ 0.5; The metabolite components that simultaneously meet conditions a) and b) are screened as differential metabolites; Use the screened differential metabolites as markers to construct a backpropagation (BP) neural network prediction model. Input the concentration data of the differential metabolites and the storage environment parameters, and output the predicted value of the shelf life of fresh-cut jackfruit; by comparing the predicted result data of the prediction model with the actual shelf life data, verify the accuracy of the marker screening, and further screen the markers based on this, and correct the prediction model.

2. The shelf life prediction method of fresh-cut jackfruit based on characteristic metabolites according to claim 1, wherein, After constructing a backpropagation (BP) neural network prediction model using the screened differential metabolites, train and optimize the model; after the training is completed, input the metabolite data of fresh-cut jackfruit collected in real time into the model, and combine the storage temperature and humidity parameters monitored in real time to dynamically adjust the predicted shelf life result; according to the dynamically adjusted predicted shelf life result, optimize the supply chain management strategy, including but not limited to adjusting the logistics transportation time, storage conditions, and sales rhythm, to ensure that fresh-cut jackfruit maintains good quality within the shelf life and reduces losses.

3. The method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites according to claim 1, wherein The GC-MS detection conditions include: Chromatographic conditions: DB-5MS capillary column 30 m × 0.25 mm × 0.25 μm, the carrier gas is high-purity helium, the constant flow rate is 1.2 mL / min, the injection port temperature is 250 °C, splitless injection, solvent delay 3.5 min; programmed temperature rise: maintain at 40 °C for 3.5 min, rise to 100 °C at a rate of 10 °C / min, then rise to 180 °C at a rate of 7 °C / min, and finally rise to 280 °C at a rate of 25 °C / min, and maintain for 5 min; Mass spectrometry conditions: electron impact ionization source (EI), ion source temperature 230 °C, quadrupole temperature 150 °C, mass spectrometry interface temperature 280 °C, electron energy 70 eV, the scanning mode is selected ion monitoring (SIM) mode, and qualitative and quantitative ions are accurately scanned.

4. The method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites according to claim 1, wherein The differential metabolite screening further includes: Plot the common differential metabolites screened into a metabolite abundance box plot to show the change trend of differential metabolites under different freshness levels.

5. The method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites according to claim 1, wherein, The input parameters of the BP neural network model include: The concentration data of differential metabolites determined by the OPLS-DA model and multiple screening criteria, including the concentration data of volatile and non-volatile metabolites with significant differences under different freshness levels; The storage temperature parameters and storage time cumulative parameters monitored in real time, where the temperature parameters are accurate to 0.1 °C and the time parameters are in hours, and the storage temperature and time parameters are recorded; Incorporate the core indicators characterizing fruit quality, including soluble solids content, measured by a refractometer with an accuracy of ±0.5° Brix; total acid content, calculated as tartaric acid, determined by acid-base titration; and pH value, with an accuracy of ±0.

01.

6. The method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites according to claim 1, wherein When screening for differential metabolites based on the OPLS-DA model, set the number of biological replicates ≥ 3; In univariate analysis, use the P-value or FDR value as the statistical significance determination index, where the P-value needs to satisfy P < 0.05, or the FDR value is controlled at ≤ 0.05, and the number of biological replicates corresponding to the univariate analysis ≥ 2.

7. The method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites according to claim 1, wherein The screening step with variable importance in projection VIP value > 1 includes: Based on ≥3 groups of independent biological replicate samples, each group containing more than 5 fresh-cut fruit bracts, calculate the coefficient of variation CV of the metabolite peak areas within the group, and only retain metabolites with CV ≤ 15% for OPLS-DA modeling; For volatile metabolites detected by GC-MS, calculate the retention time RT drift through quality control samples. For metabolites with a retention time RT deviation ≤ 0.15 min and a relative standard deviation RSD < 2%, multiply the variable importance in projection VIP value by a stability coefficient of 1.2; For metabolites with a fold change ≥ 2 or ≤ 0.5 in fresh-cut jackfruit at different freshness levels, add a differential enhancement factor of 0.3 to the variable importance in projection VIP value; Calculate the standard deviation SD of the variable importance in projection VIP values for 3 independent experiments, and only retain the variable importance in projection VIP values for metabolites with a standard deviation SD < 0.1, excluding candidate markers with poor repeatability; For metabolites with variable importance in projection VIP > 1, require the peak shape symmetry factor As to be between 0.9 - 1.1 and the signal-to-noise ratio S / N > 30 to exclude instrument noise interference; Introduce 200 permutation tests to ensure that the metabolites with variable importance projection VIP > 1 have a predictive ability Q in the permutation model 2 The intercept < 0.05 to prevent false positives caused by overfitting.

8. The method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites according to claim 1, wherein The process of placing each sample in a headspace vial after liquid nitrogen grinding and adding saturated NaCl solution and internal standard solution specifically includes: Using a cryogenic grinder with a frequency of 18 - 20 Hz, perform liquid nitrogen grinding on fresh-cut fruit bracts, control the single grinding time within 30 - 45 s, continuously supplement liquid nitrogen during the grinding process to maintain the temperature ≤ -196 °C, ensure a cell breakage rate ≥ 95%, and at the same time avoid the degradation of volatile metabolites caused by high temperature; Immediately transfer the ground sample to a pre-cooled 50 mL headspace vial at -80 °C, and the transfer time < 15 s; Add the solution according to the ratio of sample mass g : saturated NaCl solution volume mL = 1:4, utilize the salting-out effect, with an ionic strength ≥ 6 mol / L, to promote the transfer of volatile metabolites from the aqueous phase to the headspace phase; Add 20 μL of deuterated internal standard solution with a concentration of 10 μg / mL, and its retention time differs from the target metabolite by ≥ 2 min; Add 0.1 % v / v ascorbic acid to the internal standard solution.

9. The method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites according to claim 1, wherein, The HS-SPME detection conditions include: Extraction was carried out using an SPME Arrow extraction head, and the extraction conditions were controlled as follows: constant temperature oscillation at 60 °C for 5 min, oscillation rate of 200 rpm; during headspace extraction for 15 min, the extraction head was kept 5 mm from the liquid surface of the sample; desorption at 250 °C for 5 min.

10. The method for predicting the shelf life of fresh-cut jackfruit based on characteristic metabolites according to claim 1, wherein, The method for dynamically adjusting the predicted shelf life results in combination with the real-time monitored storage temperature and humidity parameters includes: Deploy a temperature and humidity sensor array in the storage environment, with a temperature accuracy of ± 0.1 °C and a humidity accuracy of ± 2 % RH. Collect data in real time at intervals of every 5 minutes, and synchronize it to the data processing terminal through the LoRa wireless transmission module to construct a three-dimensional environmental parameter grid covering the storage space, with a spatial resolution of ≤ 0.5 m 3 ; Based on the Arrhenius equation, the degradation rates of characteristic metabolites at different temperatures were fitted. When the real-time temperature deviated from the preset storage temperature by ±2 °C, the attenuation coefficient of metabolite concentration in the BP neural network was automatically adjusted, and the adjustment step size was ≤ 0.05 min; When the humidity > 85% RH, a mold growth warning was triggered, and a shelf life reduction coefficient of 0.8 - 0.9 was superimposed on the output layer of the model; when the humidity < 60% RH, a water loss rate compensation parameter of 10% - 15% was increased to correct the quality prediction deviation caused by water loss; The sliding window incremental learning algorithm was adopted. Every 24 h, the latest 50 groups of temperature and humidity data and the corresponding concentration values of differential metabolites were input into the BP neural network to locally update the weight matrix related to environmental parameters, ensuring the adaptive ability of the model to seasonal environmental fluctuations, and the prediction error was dynamically converged to ≤ 5%.

Images