Machine learning and amino acid sequence information-based umami peptide recognition threshold prediction method

Abstract

The invention provides a machine learning and amino acid sequence information-based umami peptide recognition threshold prediction method, which comprises the following steps of: obtaining umami peptide data, and preprocessing the umami peptide data; performing digital characterization on the preprocessed umami peptide data through a molecular fingerprint, a molecular descriptor and an amino acid matrix; performing feature screening on the digitally represented umami peptide data through Pearson correlation, F regression, mutual information and a wrapping mode to obtain screened molecular descriptors, amino acid counts and molecular fingerprints; sub-models are constructed through the screened molecular descriptors, amino acid counts and molecular fingerprints, evaluation is conducted through regression model evaluation indexes, first six sub-models with the best effect are selected to conduct integrated model construction, and the integrated model is fitted through a gradient elevator model; and predicting the recognition threshold value of the umami peptide through the constructed integrated model. The method can quantitatively predict a large number of umami peptides, thereby providing quantitative reference for large-scale screening of umami peptides.

Description

Technical Field

[0001] The invention belongs to the field of machine learning, and in particular relates to a method for predicting an umami peptide recognition threshold based on machine learning and amino acid sequence information. Background Art

[0002] Umami is one of the five basic tastes. Umami peptides have become the most reported umami component due to their good nutritional properties and taste sensory perception. Common umami substances include nucleotides and derivatives (30 types such as IMP and BMP), amino acids, peptides, organic acids and salts (succinic acid), etc. Among them, umami peptides are the most studied and widely distributed flavor substances. Umami peptides can be used as natural taste modifiers and nutritional supplements to reduce dependence on synthetic additives. Traditional umami peptide mining methods, such as ultrafiltration, gel filtration chromatography, reversed-phase high-performance liquid chromatography and mass spectrometry, have problems such as high labor costs, long time cycles, and low yields. With the advancement of computer performance and algorithm accuracy, computer-assisted rapid screening of umami peptides has become possible. The existing model can only perform binary classification judgments on umami or not, which can no longer meet the needs of rapid pre-screening of large quantities of umami peptides. Therefore, a high-performance umami intensity prediction model is urgently needed. Summary of the invention

[0003] In view of the deficiencies of the prior art, the present invention provides a method for predicting the recognition threshold of umami peptides based on machine learning and amino acid sequence information.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] A method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information, comprising:

[0006] Acquiring umami peptide data, and preprocessing the umami peptide data;

[0007] The preprocessed umami peptide data were digitally characterized by molecular fingerprints, molecular descriptors, and amino acid matrices;

[0008] The digitally characterized umami peptide data were subjected to feature screening by Pearson correlation, F regression, mutual information and wrapping to obtain the screened molecular descriptors, amino acid counts and molecular fingerprints.

[0009] The sub-models were constructed by using the screened molecular descriptors, amino acid counts and molecular fingerprints, and were evaluated by using the regression model evaluation index. The top 6 sub-models with the best effects were selected to construct the integrated model, which was then fitted by the gradient boosting machine model.

[0010] The umami peptide recognition threshold is predicted using the constructed integrated model.

[0011] Furthermore, the preprocessing is specifically as follows:

[0012] The acquired umami peptide data were first subjected to sensory testing and in vitro synthesis verification, then deduplication processing was performed, and finally the recognition thresholds of the same peptides were averaged.

[0013] Furthermore, the regression model evaluation indicators are determination coefficient, variance explanation rate and mean absolute error.

[0014] Furthermore, the coefficient of determination (R 2 ), explained variance (Explaine_variance) and peace

[0015] The formula for mean absolute error (MAE) is as follows:

[0016]

[0017]

[0018]

[0019] Among them, y i represents the i-th observation value; Represents the model's predicted value for the i-th observation; All y i The average value; ∑ is the summation symbol, which means that the following expression is accumulated from the first observation value to the nth observation value i; Var{y} represents the variance of the observation value, which indicates the total variation of the data; Represents the actual value y and predicted value of each observation The absolute value of the difference between .

[0020] Furthermore, the specific parameters of the Pearson screening are as follows:

[0021] The Pearson correlation coefficient measures the linear relationship between X-Input and Y-Value. The measurement result is between 1- and 1. If the result is 1 or -1, it represents a linear positive correlation and negative correlation respectively. If the result is 0, it means there is no correlation.

[0022] Furthermore, the specific parameters of the F regression are as follows:

[0023] The univariate linear regression test (F_regression) uses a fast linear model to test the individual regression effects of multiple regressors in turn, with F scores (F-statistic) and significance (p-values) as evaluation results.

[0024] Furthermore, the specific parameters of the mutual information are as follows: Mutual information is a non-negative value that measures the dependency between continuous variables, it is equal to zero if and only if the two random variables are independent, and a higher value means a higher dependency.

[0025] Furthermore, the specific parameters of the wrapper filter are as follows: In order to take into account both linear data and nonlinear features, Random Forest Regressor and Linear Regression are selected as the evaluator algorithms. If a feature is true in at least one of the two model evaluators, then the indicator will be retained.

[0026] Compared with the prior art, the present invention has the following advantages:

[0027] The present invention establishes a rapid screening model for umami peptides, which is efficient and accurate in calculation, does not require complex pre-processing and expensive equipment, and realizes rapid identification based on umami peptide sequences; this method is the first regression model to realize the recognition threshold of umami peptides, and can quantitatively predict a large number of umami peptides, thereby providing a quantitative reference for the large-scale screening of umami peptides. It is worth promoting. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the description and claims, serve to illustrate the embodiments of the invention. Where appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0029] Figure 1 It is a schematic diagram of the training process and evaluation results of the model of the present invention;

[0030] Figure 2 It is a schematic diagram of the performance of the sub-models of the model of the present invention;

[0031] Figure 3 It is a schematic diagram of the web page publishing effect of the present invention. DETAILED DESCRIPTION

[0032] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0033] The present invention is further described below with specific embodiments to illustrate the implementation process of the invention.

[0034] A method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information, comprising:

[0035] Acquiring umami peptide data, and preprocessing the umami peptide data;

[0036] The preprocessed umami peptide data were digitally characterized by molecular fingerprints, molecular descriptors, and amino acid matrices;

[0037] The digitally characterized umami peptide data were subjected to feature screening by Pearson correlation, F regression, mutual information and wrapping to obtain the screened molecular descriptors, amino acid counts and molecular fingerprints.

[0038] The sub-models were constructed by using the screened molecular descriptors, amino acid counts and molecular fingerprints, and were evaluated by using the regression model evaluation index. The top 6 sub-models with the best effects were selected to construct the integrated model, which was then fitted by the gradient boosting machine model.

[0039] The umami peptide recognition threshold is predicted using the constructed integrated model.

[0040] The pre-processing is specifically as follows:

[0041] The acquired umami peptide data were first subjected to sensory testing and in vitro synthesis verification, then deduplication processing was performed, and finally the recognition thresholds of the same peptides were averaged.

[0042] The regression model evaluation indicators are determination coefficient, variance explanation rate and mean absolute error.

[0043] The coefficient of determination (R 2 ), explained variance (Explaine_variance) and mean absolute error (MAE) are as follows:

[0044]

[0045]

[0046]

[0047] (1)coefficient ofdetermination(R 2 ): Refer to Formula 1, the best score for this metric is 1.0. If the model always predicts the average y regardless of the changes in the input features, then R2 will be 0; (2) Explained variance: Refer to Formula 2, the best possible score is 1.0, lower values ​​are worse; (3) Mean absolute error: Refer to Formula 3, the function calculates the mean absolute error. The meanings of the parameters are as follows: y i represents the i-th observation value; Represents the model's predicted value for the i-th observation; All y i The average value; ∑ is the summation symbol, which means that the following expression is accumulated from the first observation value to the nth observation value i; Var{y} represents the variance of the observation value, which indicates the total variation of the data; Represents the actual value y and predicted value of each observation The absolute value of the difference between .

[0048] The specific parameters of the Pearson screening are as follows:

[0049] The Pearson correlation coefficient measures the linear relationship between X-Input and Y-Value. The measurement result is between 1- and 1. If the result is 1 or -1, it represents a linear positive correlation and negative correlation respectively. If the result is 0, it means there is no correlation.

[0050] The specific parameters of the F regression are as follows:

[0051] The univariate linear regression test (F_regression) uses a fast linear model to test the individual regression effects of multiple regressors in turn, with F scores (F-statistic) and significance (p-values) as evaluation results.

[0052] Furthermore, the specific parameters of the mutual information are as follows: Mutual information is a non-negative value that measures the dependency between continuous variables, it is equal to zero if and only if the two random variables are independent, and a higher value means a higher dependency.

[0053] Furthermore, the specific parameters of the wrapper filter are as follows: In order to take into account both linear data and nonlinear features, Random Forest Regressor and Linear Regression are selected as the evaluator algorithms. If a feature is true in at least one of the two model evaluators, then the indicator will be retained.

[0054] Table 1 is the model training data, downloaded from TastePeptidesDB (accessible at http: / / tastepeptides-meta.com / TastePeptidesDB, as of September 20, 2023), which collected 473 umami peptides, of which 426 were verified by sensory testing and in vitro synthesis. After deduplication and averaging the recognition thresholds of the same peptides from multiple studies, a total of 324 peptides with recognition thresholds were finally obtained. These umami peptides and recognition thresholds are shown in Table 1.

[0055] Table 1 Umami peptides and recognition thresholds

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070] Table 2 describes the 13 different regressor algorithms used to fit and train the models. 2 The modeling results of 13 algorithms on various feature data were compared horizontally using two indicators: variance explanation rate. Among them, decision tree regressor, additional decision tree regressor, random forest regressor, adaptive boosting regressor and gradient boosting tree regressor showed the best performance. The parameters of the sub-model with the highest score were further optimized by grid search. The parameters and performance diagram of the model are shown in Figure 2 , the test set R of these models 2 Most of them are around 0.65. The above six models are further fitted by the gradient boosting tree algorithm, and finally the integrated model UmamiIP is obtained. UmamiIP shows similar performance in the evaluation of the test set and the training set, and the absolute coefficient R 2 The variance is 84.95% / 84.46% (Train / Test set), and the explained variance is 86.82% / 84.47% (Train / Test set). This indicates that the model is not overfitted ( Figure 1 B). In contrast, the average of the predictions of these sub-models ( Figure 1 C) did not perform as well as UmamiIP. Since UmamiIP is the first tool to predict umami intensity, there is almost no similar tool for comparison, so the present invention can only compare UmamiIP with its own sub-model. UmamiIP performed much more rigorously and achieved the first place in all three indicators ( Figure 1 D). In order to better explore the prediction effect of UmamiIP, the present invention further analyzes the amino acid length distribution ( Figure 1 E) and threshold distribution ( Figure 1 F) The mean absolute error was calculated from two perspectives. From the perspective of amino acid length, the error of short peptides (1-3 amino acids in length) is relatively large. It is speculated that this is because short peptides have greater activity, so the error value is larger at the same error rate. UmamiIP has a better prediction effect on peptides with a length of 4-10, and the number of amino acids of this length in TastePeptidesDB is the largest. The data diversity used to train UmamiIP increases with the abundance of peptides of the same length. With the help of higher data diversity, UmamiIP can better predict the recognition threshold of peptides of this length. In terms of thresholds, UmamiIP works better in areas with higher threshold density (0-3mM), and then the error increases with the increase of the threshold value.

[0071] Table 2 Detailed information on hyperparameter search for 13 different regressors

[0072]

[0073]

[0074]

[0075]

[0076] Example 2

[0077] Figure 3 The graphical interface and operation logic of the web-based access program for umami peptide prediction are shown. Visitors only need to submit the amino acid sequence of the peptide to be predicted in FASTA or CSV file format to make the prediction. After three processing steps, the umami judgment results can be downloaded.

[0078] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information, characterized in that: include: Acquiring umami peptide data, and preprocessing the umami peptide data; The preprocessed umami peptide data were digitally characterized by molecular fingerprints, molecular descriptors, and amino acid matrices; The digitally characterized umami peptide data were subjected to feature screening by Pearson correlation, F regression, mutual information and wrapping to obtain the screened molecular descriptors, amino acid counts and molecular fingerprints. The sub-models were constructed by using the screened molecular descriptors, amino acid counts and molecular fingerprints, and were evaluated by using the regression model evaluation index. The top 6 sub-models with the best effects were selected to construct the integrated model, which was then fitted by the gradient boosting machine model. The umami peptide recognition threshold is predicted using the constructed integrated model.

2. The method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information according to claim 1, characterized in that: The pre-processing is specifically as follows: The acquired umami peptide data were first subjected to sensory testing and in vitro synthesis verification, then deduplication processing was performed, and finally the recognition thresholds of the same peptides were averaged.

3. The method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information according to claim 1, characterized in that: The regression model evaluation indicators are determination coefficient, variance explanation rate and mean absolute error.

4. The method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information according to claim 3, characterized in that: The coefficient of determination R 2 The formulas for , variance explanation rate and mean absolute error MAE are as follows: Among them, y i represents the i-th observation value; Represents the model's predicted value for the i-th observation; All y i The average value; ∑ is the summation symbol, which means that the following expression is accumulated from the first observation value to the nth observation value i; Var{y} represents the variance of the observation value, which indicates the total variation of the data; Represents the actual value y and predicted value of each observation The absolute value of the difference between .

5. The method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information according to claim 1, characterized in that: The specific parameters of the Pearson screening are as follows: The Pearson correlation coefficient measures the linear relationship between X-Input and Y-Value. The measurement result is between 1- and 1. If the result is 1 or -1, it represents a linear positive correlation and negative correlation respectively. If the result is 0, it means there is no correlation.

6. The method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information according to claim 1, characterized in that: The specific parameters of the F regression are as follows: The univariate linear regression test uses a fast linear model to test the individual regression effects of multiple regressors in turn, with the F score and significance as the evaluation results.

7. The method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information according to claim 1, characterized in that: The specific parameters of the mutual information are as follows: Mutual information is a non-negative value that measures the dependency between continuous variables, it is equal to zero if and only if the two random variables are independent, and a higher value means a higher dependency.

8. The method for predicting umami peptide recognition threshold based on machine learning and amino acid sequence information according to claim 1, characterized in that: The specific parameters of the wrapper filter are as follows: In order to take into account both linear data and nonlinear features, random forest regressor and linear regressor are selected as the evaluator algorithms. If a feature is true in at least one of the two model evaluators, then the indicator will be retained.

Images