A probiotic product stability prediction model, a construction method thereof and an application thereof

Abstract

The present application relates to a probiotic product stability prediction model and its construction method and application. The present application provides a new method for predicting the stability of probiotic products earlier, faster and more accurately, taking into account up to 14 potential influencing factors such as strain physiological characteristics, preparation process and environmental factors, and using machine learning algorithms to capture the complex nonlinear relationship between them and stability, greatly improving the accuracy and reliability of the prediction model, and the timeliness of the prediction is strong, the universality and guidance are strong, and digital management can be realized.

Description

Technical Field

[0001] This invention belongs to the field of bioinformatics technology, and relates to a probiotic product stability prediction model, its construction method, and its application. Background Technology

[0002] Probiotics (such as Bifidobacteria) are a class of live microorganisms beneficial to the human body, primarily colonizing the gut and maintaining health by balancing gut flora, enhancing immunity, and promoting digestion and absorption. As one of the core probiotics in the human gut, the activity and stability of probiotic products directly affect their probiotic functions. Product stability is the result of a complex interplay of multiple factors. Traditional prediction methods often rely on accelerated stability testing and calculations based on the Arrhenius equation. However, this method assumes that the reaction rate is only temperature-dependent, neglecting the significant influence of intrinsic factors such as formulation, processing, and strain characteristics.

[0003] In existing technologies, such as CN114384105A, a method is provided to optimize the prediction model by introducing water activity (Aw) and strain coefficient (S). The prediction model formula is: ln(ΔV) = a + bln(t) + c(1 / T) + dAw + eS. This method represents a significant improvement over traditional methods, but it still has the following limitations:

[0004] Model limitations: This model is essentially still a linear or log-linear model, which makes it difficult to accurately capture and fit the complex nonlinear and interactive relationships between multiple influencing factors and stability;

[0005] Insufficient feature dimensions: Only four variables were considered: time, temperature, water activity, and strain coefficient. Other key factors that have been proven to have a significant impact on stability, such as formulation process (encapsulation rate, residual oxygen content) and inherent physiological characteristics of the strain (membrane fatty acid composition, self-aggregation), were not included.

[0006] Efficiency and cost: Obtaining the strain coefficient S still requires several months (e.g., 3-6 months) of accelerated testing, making it impossible to achieve true "early" and "non-destructive" prediction. The research and development cycle and cost remain high.

[0007] In summary, there is an urgent need in this field for a new method that can predict the stability of Bifidobacterium products earlier, faster, and more accurately. Summary of the Invention

[0008] To address the shortcomings of existing technologies and practical needs, this invention provides a probiotic product stability prediction model, its construction method, and its application. It also provides a probiotic product stability prediction scheme based on multi-feature fusion, aiming to achieve high-precision prediction of long-term stability based solely on initial product attributes or short-term experimental data.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] In a first aspect, the present invention provides a method for constructing a stability prediction model for probiotic products, the method comprising the following steps:

[0011] S1. Construct a stability database: Take probiotic products with different formulations and conduct stability tests; for each product, determine its initial characteristic parameters in the initial state or at the beginning of the stability test, and use them as the initial characteristic parameter set, denoted as (X1,X2,……,Xm), where m is any positive integer. In the stability test, the number of viable bacteria in the product is periodically detected at specified times. The decay rate constant k of each product is obtained by fitting the exponential decay model N(t) = N(0)×e^(-k×t), which is used as the stability characterization value. Here, N(t) is the number of viable bacteria at the specified time, N(0) is the initial number of viable bacteria, and t is the specified time.

[0012] S2. Screening the key feature set: Based on the stability database constructed in step S1, the influence weight (importance score) of the initial feature parameter set on the decay rate constant k is analyzed using machine learning algorithms, and n key feature parameters are screened out (they can be sorted from largest to smallest according to their influence weight, and the first n feature parameters are screened, n < m) to form the key feature set;

[0013] S3. Training the machine learning prediction model: Using the parameter values ​​of the key feature set as input variables and the decay rate constant k as output variables, the machine learning algorithm is trained to obtain a probiotic product stability prediction model.

[0014] This invention designs a multi-feature fusion model for predicting the stability of probiotic (such as Bifidobacterium) products. By introducing a more comprehensive feature system and using machine learning algorithms to mine the complex mapping relationship between features and stability, it aims to achieve high-precision prediction of long-term stability based solely on the initial attributes of the product or short-term experimental data.

[0015] Preferably, the initial characteristic parameters in step S1 include water activity, encapsulation rate, strain type code, residual oxygen content, storage temperature, proportion of unsaturated fatty acids in cell membranes, pH value, content of protective agent, barrier properties of packaging material, type and content of sugar alcohol, relative humidity, antioxidant content, survival rate after freeze-drying, and cell self-aggregation.

[0016] Preferably, the stability test in step S1 includes accelerated stability test and room temperature stability test.

[0017] Preferably, the accelerated stability test includes storing the probiotic product at 30℃~50℃, preferably 35℃~45℃, and more preferably 37℃ ± 2℃.

[0018] Preferably, the room temperature stability test includes storing the probiotic product at 5℃~40℃, preferably 15℃~30℃, and more preferably 25℃ ± 2℃.

[0019] Preferably, the machine learning algorithm in step S2 includes random forest or gradient boosting decision tree algorithm.

[0020] Preferably, the key characteristic parameters in step S2 include water activity, encapsulation rate, strain type code, residual oxygen content, storage temperature, proportion of unsaturated fatty acids in cell membrane, pH value, and content of protective agent.

[0021] Preferably, the machine learning algorithm described in step S3 includes any one of gradient boosting decision tree (XGBoost, LightGBM), random forest, or support vector machine (SVR).

[0022] Secondly, the present invention provides a probiotic product stability prediction model, which is constructed by the method for constructing a probiotic product stability prediction model described in the first aspect.

[0023] Thirdly, the present invention provides a method for predicting the stability of probiotic products. The prediction method includes: obtaining key characteristic parameters of the probiotic product to be predicted, including water activity, encapsulation rate, strain type code, residual oxygen content, storage temperature, proportion of unsaturated fatty acids in cell membranes, pH value, and content of preservative; inputting the key characteristic parameters into the probiotic product stability prediction model described in the first aspect, and outputting a predicted decay rate constant k_pred; and calculating the predicted number of viable bacteria and survival rate of the product after a specified storage time based on the decay rate constant k_pred and the initial viable bacteria count of the product to be predicted.

[0024] Preferably, the formula for calculating the number of viable bacteria is: N(t) = N(0)×e^(-k_pred×t), and the formula for calculating the survival rate is: survival rate = N(t) / N(0)×100%.

[0025] Thirdly, the present invention provides a system for predicting the stability of probiotic products, the system being used to perform the steps of the prediction method described in the second aspect; the prediction system includes:

[0026] Data acquisition module: used to acquire or input key characteristic parameters of the probiotic product to be tested;

[0027] Model storage module: used to store probiotic product stability prediction models;

[0028] Prediction Calculation Module: Calls the prediction model in the model storage module, calculates the feature parameters of the data acquisition module, and outputs the predicted decay rate constant k_pred;

[0029] The results output module is used to display the predicted decay rate constant k_pred, and can further calculate and display the predicted number of viable bacteria and survival rate after a specified storage time based on the k_pred value.

[0030] Compared with the prior art, the present invention has at least the following beneficial effects:

[0031] This invention provides a multi-feature fusion-based probiotic product stability prediction scheme. It incorporates up to 14 potential influencing factors, including strain physiological characteristics, formulation processes, and environmental factors, and utilizes machine learning algorithms to capture their complex nonlinear relationships with stability, significantly improving the accuracy and reliability of the prediction model. The scheme boasts strong predictive timeliness: only initial characteristic parameters (such as Aw, encapsulation rate, and residual oxygen content) after product manufacturing are required, which can be measured within hours, to immediately predict long-term stability, eliminating the need to wait for lengthy accelerated testing results. This shortens the prediction cycle from months to hours, significantly improving R&D efficiency and reducing development costs. Furthermore, the model offers strong versatility and guidance: it not only makes predictions but also reveals key drivers affecting stability through feature importance ranking, providing clear data support and directional guidance for formulation optimization and process improvement (such as how to adjust encapsulation rate or Aw to achieve the target shelf life). Finally, it enables digital management: it can be integrated into an enterprise's quality management platform to automate, intelligently and visually predict stability, promoting the digital upgrade of the product quality management system. Attached Figure Description

[0032] Figure 1 The flowchart below shows the overall process of the stability prediction scheme provided in the embodiments of the present invention.

[0033] Figure 2 This is a schematic diagram illustrating the ranking of feature importance in an embodiment of the present invention.

[0034] Figure 3 This is a scatter plot of the model's predicted values ​​and the actual values ​​in an embodiment of the present invention. Detailed Implementation

[0035] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments. However, the following examples are merely simplified examples of the present invention and do not represent or limit the scope of protection of the present invention. The scope of protection of the present invention is determined by the claims.

[0036] In the examples, where no specific techniques or conditions are indicated, the procedures shall be followed according to the techniques or conditions described in the literature in the art, or according to the product instructions. Where the manufacturers of the reagents or instruments are not indicated, they are all conventional products that can be purchased through regular commercial channels.

[0037] The present invention designs a probiotic product stability prediction scheme based on multi-feature fusion, and in specific examples, a method for constructing a probiotic product stability prediction model and a model-based prediction method can be provided. The schematic flow chart is shown in Figure 1 , the construction method of the prediction model comprises the following steps:

[0038] S1. Constructing a stability database: taking probiotic products with different formulas and performing stability tests; for each product, measuring its initial characteristic parameters in the initial state or at the early stage of the stability test, which is recorded as the initial characteristic parameter set, denoted as (X1, X2, ..., Xm), where m is any positive integer, and regularly detecting the number of viable bacteria in the product at a specified time during the stability test, fitting to obtain the decay rate constant k of each product based on the exponential decay model N(t) = N(0)×e^(-k×t), which is used as the stability characterization value, wherein N(t) is the number of viable bacteria at the specified time, N(0) is the initial number of viable bacteria, and t is the specified time;

[0039] Further, the initial characteristic parameters in step S1 include water activity, embedding rate, strain type code, residual oxygen content, storage temperature, proportion of unsaturated fatty acids in cell membrane, pH value, content of protective agent, barrier property of packaging material, type and content of sugar alcohol, relative environmental humidity, content of antioxidant, survival rate after freeze-drying and cell auto-aggregation;

[0040] S2. Screening a key feature set: based on the stability database constructed in step S1, analyzing the influence weight of the initial characteristic parameter set on the decay rate constant k by using a machine learning algorithm, and screening n key characteristic parameters (n <m) from the parameters to form a key feature set;

[0041] Further, the key characteristic parameters in step S2 include water activity, embedding rate, strain type code, residual oxygen content, storage temperature, proportion of unsaturated fatty acids in cell membrane, pH value and content of protective agent;

[0042] S3. Training a machine learning prediction model: taking the parameter values of the key feature set as input variables and taking the decay rate constant k as an output variable, training a machine learning algorithm to obtain a probiotic product stability prediction model.

[0043] The prediction method includes: obtaining key characteristic parameters of the probiotic product to be predicted, including water activity, encapsulation rate, strain type code, residual oxygen content, storage temperature, proportion of unsaturated fatty acids in cell membranes, pH value, and content of preservative; inputting the key characteristic parameters into the probiotic product stability prediction model and outputting the predicted decay rate constant k_pred; and calculating the predicted viable count and survival rate of the product after a specified storage time based on the decay rate constant k_pred and the initial viable count of the product to be predicted.

[0044] In another embodiment of the present invention, a prediction system for the stability of probiotic products may also be provided. The prediction system is used to perform the steps in the prediction method described above, and the prediction system includes:

[0045] Data acquisition module: used to acquire or input key characteristic parameters of the probiotic product to be tested;

[0046] Model storage module: used to store probiotic product stability prediction models;

[0047] Prediction Calculation Module: Calls the prediction model in the model storage module, calculates the feature parameters of the data acquisition module, and outputs the predicted decay rate constant k_pred;

[0048] The results output module is used to display the predicted decay rate constant k_pred, and can further calculate and display the predicted number of viable bacteria and survival rate after a specified storage time based on the k_pred value.

[0049] Example 1

[0050] This embodiment constructs a prediction model.

[0051] Sample Preparation and Data Acquisition: One hundred different formulations of Bifidobacterium lyophilized powder samples (combinations of one to three lyophilized powders) were prepared, with strains including *Bifidobacterium animalis* subsp. *lactolaccos* BLa80, *Bifidobacterium longum* subsp. *longum* BL21, and *Bifidobacterium longum* subsp. *infantum* BI45. Fourteen initial characteristic parameters of each sample were measured using appropriate instruments (ranges are shown in Table 1). Simultaneously, all samples underwent accelerated stability testing at 37℃ and 60% ± 15% relative humidity, with samples taken at the end of 0, 1, 2, 3, 4, 5, and 6 months to determine viable cell counts. Room temperature stability testing was conducted at 25℃ and 60% ± 15% relative humidity, with samples taken at the end of 0, 3, 6, 9, 12, 18, and 24 months to determine viable cell counts.

[0052] Calculate the decay rate constant k of the output variable: For each sample, perform a linear regression of the logarithm of its viable cell count (lnCFU / g) against time (months), and the slope of the regression line is the decay rate constant k.

[0053] Feature selection and model training dataset construction: For each probiotic product sample, we defined its true stability index k_ture as the decay rate constant obtained by fitting data from at least 24 months of long-term stability tests at room temperature (T2 = 25℃ ± 2℃). We collected 14 initial feature parameters measured in the initial state of these 100 samples, along with the corresponding k_ture value for each sample, which together constituted an initial dataset for model training containing 100 data points. The XGBoost algorithm was used to analyze the importance of the 14 initial features in predicting k_ture, and the top 8 most important features (such as...) were selected. Figure 2 As shown in the figure, the parameters are: water activity (Aw), embedding rate, strain type code, residual oxygen, storage temperature (T), proportion of unsaturated fatty acids in the cell membrane, pH value, and protective agent content. To improve the robustness of the model, in subsequent model training, we also added short-term data (eigenvalues ​​and the decay rate constant k_accelerated from the accelerated stability test) obtained from some samples under accelerated stability test (T1 = 37℃ ± 2℃) as supplementary data to the training process. However, the evaluation and validation of the model are always based on the prediction accuracy of k_ture.

[0054] Model Training and Validation: Using these 8 key features as input and the decay rate constant as output, the dataset was randomly divided into a training set (80 samples) and a test set (20 samples) in an 8:2 ratio. The XGBoost model was trained using the training set and validated using the test set. Results are as follows: Figure 3 As shown, the model's predicted values ​​are in high agreement with the actual values, with a coefficient of determination R² > 0.98 and a root mean square error RMSE < 0.02, proving that the model has extremely high accuracy.

[0055] Table 1. Key factors affecting the stability of probiotics (14 in total)

[0056]

[0057]

[0058] Example 2

[0059] To verify the accuracy and reliability of the stability prediction model constructed in this invention, we designed the following verification experiment to compare the model prediction results with the actual measurement results of the long-term stability test, which is recognized as the "gold standard" in the field.

[0060] Validation methods and materials: A new strain of Bifidobacterium breve BBr96 (strain number 11) was introduced to prepare freeze-dried bacterial powder samples with different formulations. Three new Bifidobacterium products that did not participate in model training were randomly selected, and their formulations were different from those of the samples used in model training.

[0061] Verification Method: Immediately after the production of the verification samples, eight key characteristic parameters were measured. These parameters were input into the prediction model trained in Example 1, which output the predicted decay rate constant k_pred. This model was then used to calculate the viable cell survival rate (N(t) = N(0)×e^(-k_pred×t), survival rate = N(t) / N(0)×100%) after storage at 25℃ (298K) for 12 and 18 months. The same batch of verification samples was then stored under long-term stability test conditions of 25℃ ± 2℃ and 60% ± 15% relative humidity. Samples were taken at 12 and 18 months of storage, and the total viable cell count was determined according to national standard methods. The actual viable cell survival rate was then calculated.

[0062] The validation sample A had the following characteristics: Aw: 0.16, encapsulation rate: 96%, strain code: 1811 (BLa80 / BI45 / BBr96), O2: 0.8%, T: 298K, UFA ratio: 0.58, pH: 4.4, and protectant content: 18%.

[0063] Validation sample B: Aw: 0.28, embedding rate: 75%, strain code: 2411 (BLa36 / BL11 / BBr96), O2: 2.5%, T: 298K, UFA ratio: 0.61, pH: 4.1, protectant content: 8%;

[0064] Validation sample C: Aw: 0.13, embedding rate: 92%, strain code: 91011 (BI03 / WKB148 / BBr96), O2: 1.0%, T: 298K, UFA ratio: 0.55, pH: 4.5, protectant content: 22%.

[0065] Table 2. Validation results of the prediction model accuracy

[0066]

[0067] Furthermore, the prediction results of the model of this invention are compared with the prediction results using the linear model described in CN114384105A. Using the same verification sample feature data, predictions are made using both models, and the absolute errors are compared with the measured results of long-term experiments.

[0068] Table 3 Comparison of prediction accuracy with existing technologies (linear models)

[0069]

[0070] The predicted values ​​of the model for all validation samples at 12 and 18 months are in high agreement with the measured values ​​from long-term experiments. The absolute error of all prediction results is less than 2.5 percentage points, which fully demonstrates that the prediction model of this invention has extremely high accuracy.

[0071] In contrast, the existing linear model (CN114384105A) has a significantly larger prediction error (up to 8.9 percentage points). This indicates that the present invention, by introducing more key features and employing machine learning algorithms to capture nonlinear relationships, has achieved a significant and substantial improvement in prediction accuracy compared to the traditional linear model.

[0072] In summary, this invention provides a novel method for predicting the stability of probiotic products earlier, faster, and more accurately. It incorporates up to 14 potential influencing factors, including strain physiological characteristics, formulation processes, and environmental factors, and utilizes machine learning algorithms to capture their complex nonlinear relationships with stability. This significantly improves the accuracy and reliability of the prediction model, offering strong predictive timeliness, versatility, and guidance, enabling digital management. Rigorous verification experiments demonstrate that the stability prediction method provided by this invention can accurately predict the long-term storage stability of new products during the R&D stage, based solely on initial parameters. The prediction error is small, far superior to existing technologies, and it possesses outstanding practical value and promising industrial application prospects.

[0073] The applicant declares that the above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention fall within the protection and disclosure scope of the present invention.

Claims

1. A method for constructing a stability prediction model for probiotic products, characterized in that, The method includes the following steps: S1. Constructing a stability database: Probiotic products with different formulations were tested for stability. For each product, its initial characteristic parameters in the initial state or at the beginning of the stability test were measured and used as the initial characteristic parameter set, denoted as (X1, X2, ..., Xm), where m is any positive integer. The viable cell count of the product was periodically detected at specified times during the stability test. The decay rate constant k of each product was obtained by fitting the exponential decay model N(t) = N(0) × e^(-k × t) as the stability characterization value. Here, N(t) is the viable cell count at the specified time, N(0) is the initial viable cell count, and t is the specified time. The initial characteristic parameters include water activity, encapsulation efficiency, strain type code, residual oxygen, storage temperature, proportion of unsaturated fatty acids in cell membranes, pH value, content of preservatives, barrier properties of packaging materials, type and content of sugar alcohols, relative humidity, content of antioxidants, survival rate after freeze-drying, and cell self-aggregation. S2. Screening of key feature set: Based on the stability database constructed in step S1, the influence weight of the initial feature parameter set on the decay rate constant k is analyzed using machine learning algorithms, and n key feature parameters are screened out to form a key feature set; the key feature parameters include water activity, encapsulation rate, strain type code, oxygen residue, storage temperature, cell membrane unsaturated fatty acid ratio, pH value and protectant content; S3. Training the machine learning prediction model: Using the parameter values ​​of the key feature set as input variables and the decay rate constant k as output variables, the machine learning algorithm is trained to obtain a probiotic product stability prediction model.

2. The method for constructing a probiotic product stability prediction model according to claim 1, characterized in that, The stability test described in step S1 includes accelerated stability test and room temperature stability test.

3. The method for constructing a probiotic product stability prediction model according to claim 2, characterized in that, The accelerated stability test includes storing the probiotic product at 37℃ ± 2℃.

4. The method for constructing a probiotic product stability prediction model according to claim 2, characterized in that, The room temperature stability test includes storing the probiotic product at 25℃ ± 2℃.

5. The method for constructing a probiotic product stability prediction model according to claim 1, characterized in that, The machine learning algorithm described in step S2 includes random forest or gradient boosting decision tree.

6. The method for constructing a probiotic product stability prediction model according to claim 1, characterized in that, The machine learning algorithm mentioned in step S3 includes any one of gradient boosting decision tree, random forest or support vector machine.

7. A method for predicting the stability of probiotic products, characterized in that, The prediction method includes: Obtain key characteristic parameters of the probiotic product to be predicted, including water activity, encapsulation rate, strain type code, residual oxygen content, storage temperature, proportion of unsaturated fatty acids in cell membrane, pH value, and content of protective agent; The key feature parameters are input into the probiotic product stability prediction model, and the predicted decay rate constant k_pred is output; the probiotic product stability prediction model is constructed by the method for constructing a probiotic product stability prediction model as described in any one of claims 1-6; The predicted viable count and survival rate of the product after a specified storage time are calculated based on the decay rate constant k_pred and the initial viable count of the product to be predicted.

8. The prediction method according to claim 7, characterized in that, The formula for calculating the number of viable bacteria is: N(t) = N(0)×e^(-k_pred×t), and the formula for calculating the survival rate is: survival rate = N(t) / N(0)×100%.

9. A system for predicting the stability of probiotic products, characterized in that, The prediction system is used to perform the steps in the prediction method of claim 7 or 8; the prediction system includes: Data acquisition module: used to acquire or input key characteristic parameters of the probiotic product to be tested; Model storage module: Used to store probiotic product stability prediction models; Prediction Calculation Module: Calls the prediction model in the model storage module, calculates the feature parameters of the data acquisition module, and outputs the predicted decay rate constant k_pred; The results output module is used to display the predicted decay rate constant k_pred, and can calculate and display the predicted number of viable bacteria and survival rate after a specified storage time based on the k_pred value.

Images