Novel sugar chain structure identification method and device based on GU value

Abstract

The invention relates to the technical field of bioinformatics, and discloses a novel sugar chain structure identification method based on a GU value, and the method comprises the following steps: calculating the GU value of an unknown sugar chain sample; converting the known carbohydrate chain structure into a known carbohydrate chain feature vector; selecting a graph convolutional network model to process the sequence, and training a sugar chain structure model by using a cross validation method; inputting the GU value of the sugar chain sample into the model for similarity analysis; compared with a sugar chain structure identification method depending on accurate molecular weight or sequence matching in the prior art, efficient and accurate identification is difficult to realize when a complex sugar chain structure is processed, especially under the condition that a sample structure is complex. According to the method, GU value calculation is combined with the deep learning model to predict the carbohydrate chain structure, so that the structure of an unknown carbohydrate chain can be efficiently and accurately deduced, the limitation caused by high complexity and long consumed time in the prior art is avoided, and the efficiency and accuracy of carbohydrate chain structure identification are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and in particular to a novel sugar chain structure identification method based on GU value. Background Art

[0002] Sugar chain structures play a vital role in many biological processes, including cell recognition, signal transduction, and immune response. However, due to the complexity and diversity of sugar chain structures, accurate identification of them has always been a challenge in bioinformatics. Traditionally, mass spectrometry (MS) and nuclear magnetic resonance spectroscopy (NMR) techniques are the main means of identifying sugar chain structures, which can provide accurate molecular weight and partial structural information. However, mass spectrometry and NMR techniques often rely on accurate molecular weight or sequence matching to identify sugar chain structures. For some highly complex sugar chain structures, especially those with slight differences in molecular weight or sequence, this method has insufficient discrimination and is difficult to provide efficient and accurate identification results. In addition, the experimental process of traditional methods is complicated, the analysis time is long, and a large amount of sample processing and data analysis are usually required, resulting in the entire identification process being time-consuming and laborious. Secondly, the existing sugar chain structure database retrieval methods also have limitations. These methods are usually based on direct matching of molecular weight or sequence, but lack effective tools to deal with the diversity of sugar chain structures caused by modification, connection mode or spatial conformation. In particular, in the case of complex sugar chain sample structures, existing technologies are difficult to achieve rapid structural inference and matching, which seriously affects the efficiency of sugar chain structure identification. Therefore, a new method is urgently needed to provide more efficient and accurate identification means in the context of complex sugar chain structures to address the shortcomings of existing technologies. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to propose a new sugar chain structure identification method based on GU value, which aims to overcome the limitations of traditional methods in identification efficiency and accuracy by introducing GU value calculation and deep learning model, and realize efficient and accurate inference and identification of unknown sugar chain structures.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solution: The present invention provides a novel sugar chain structure identification method based on GU value,

[0005] Preferably, the step S10 comprises calibrating the migration time of the standard sample using an internal standard;

[0006] Preferably, the structural information of the known sugar chain in step S10 includes glycosidic bond type, monosaccharide unit, modification group, spatial conformation and molecular weight information;

[0007] Preferably, the data cleaning step in step S20 includes removing redundant data, filling missing values, and standardizing sugar chain structure information;

[0008] Preferably, in step S30, the generation of the feature vector includes encoding the sugar chain structure using a molecular fingerprint ECFP;

[0009] Preferably, according to the novel sugar chain structure identification method based on GU value of claim 1, it is characterized in that the step S40 further comprises using L2 regularization to prevent model overfitting;

[0010] Preferably, the novel sugar chain structure identification device based on GU value comprises:

[0011] A GU value calculation module is used to obtain the migration time of the unknown sugar chain sample through a capillary electrophoresis experiment, and calculate the GU value of the unknown sugar chain sample in combination with the migration time of the adjacent standard sample;

[0012] A data collection and processing module is used to collect the structural information of known sugar chains and calculate the corresponding GU values, organize the structural information and the corresponding GU values ​​into a data set, and represent the known sugar chain structure in the form of SMILES after data cleaning;

[0013] A feature vector generation module is used to convert the known sugar chain structure into a known sugar chain feature vector, and use the known GU value as the target variable of the sugar chain structure model;

[0014] a model training module, for arranging sugar chain structure feature vectors to obtain a sugar chain structure feature sequence, selecting a graph convolutional network model to process the sequence, and training a sugar chain structure model using a cross-validation method;

[0015] The similarity analysis module is used to input the calculated GU value of the unknown sugar chain sample into the pre-trained sugar chain structure model to perform sugar chain structure similarity analysis.

[0016] The beneficial effect of the present invention is that compared with the sugar chain structure identification method in the prior art that relies on accurate molecular weight or sequence matching, it is difficult to achieve efficient and accurate identification when dealing with complex sugar chain structures, especially under the condition of complex sample structures. The present invention predicts the sugar chain structure by using GU value calculation combined with a deep learning model, so that the structure of unknown sugar chains can be efficiently and accurately inferred, avoiding the limitations of the prior art due to high complexity and long time consumption, and significantly improving the efficiency and accuracy of sugar chain structure identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 Schematic diagram of the process of the first embodiment of the novel sugar chain structure identification method based on GU value of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Embodiment 1: Figure 1 FIG. 1 is a schematic diagram of the process of the first embodiment of the novel sugar chain structure identification method based on GU value of the present invention.

[0021] Step S10: obtaining the migration time of the unknown sugar chain sample by capillary electrophoresis experiment, and calculating the GU value of the unknown sugar chain sample in combination with the migration time of the adjacent standard sample;

[0022] It should be noted that capillary electrophoresis experiment refers to the process of separating and analyzing sugar chain samples in a capillary. The experiment uses an electric field to make sugar chain molecules migrate in the capillary. According to the differences in molecular weight, shape or charge, the sugar chain samples will migrate at different speeds in the capillary, thereby achieving the purpose of separation. By accurately measuring the migration time of the sugar chain samples, it can provide necessary data support for the subsequent GU value calculation.

[0023] It is understandable that in multiple groups of experiments, the migration time ratio between ladders is constant, that is, for a given ladder, its migration time can be expressed as: T n -T n-1 =k n ×(T2-T1)

[0024] Among them, T1, T2, ...T n is the migration time of the known ladder, k n is the time migration ratio. In a set of experiments, k n is fixed, so the virtual ladder position of the sample can be inferred from the ladder migration time of the same group.

[0025] It should be understood that the relationship between the GU value of a sugar chain and its migration time is linear between adjacent ladders, so the GU value of an unknown sugar chain can be inferred from the migration time of adjacent ladders. That is:

[0026] Where GUx is the GU value of the unknown sugar chain X, G N is the GU value of the virtual oligomaltosaccharide ladder closest to the unknown sugar chain, T N , T N+1 It is the migration time of the Nth and N+1th ladders.

[0027] Step S20: collecting structural information of known sugar chains and calculating corresponding GU values, organizing the structural information and corresponding GU values ​​into a data set, and obtaining the known sugar chain structure after data cleaning, which is expressed in the form of SMILES;

[0028] It should be noted that the structural information of known sugar chains includes but is not limited to the following: glycosidic bond type, monosaccharide unit, modification group, spatial conformation and molecular weight information. These information are the core elements of sugar chain structure. By comprehensively collecting these elements, the integrity and accuracy of the data set can be ensured. In addition, the calculated GU value is based on the experimental data of known sugar chains and is associated with the structural information to form a correspondence between structure and function. Data cleaning is a crucial part of the data preparation process, including removing redundant data, filling missing values ​​and standardizing sugar chain structural information. Removing redundant data means deleting duplicate or invalid records to improve the quality and availability of the data set. Filling missing values ​​is to supplement the missing structural information through reasonable inference or data interpolation methods to avoid the problem of incomplete data during model training. Standardized sugar chain structural information is to convert various forms of sugar chain data into standardized representations, such as using the SMILES format, for subsequent model processing, where the SMILES format is a simplified molecular structure representation method that represents the chemical structure of sugar chains through linear symbols so that it can be efficiently processed and parsed in computer systems. This format is not only concise, but also compatible with most bioinformatics tools and is suitable for processing and analysis of large-scale data sets.

[0029] It is understandable that the goal of data cleaning is to improve the quality of the data set so that subsequent model training can be carried out on the basis of high-quality data, thereby improving the accuracy and robustness of the model.

[0030] It should be understood that the specific implementation steps of data cleaning can be:

[0031] (1) Remove redundant data

[0032] For example, a sugar chain structure database contains 10,000 records, in which a sugar chain structure A has multiple duplicate records. The record example is:

[0033] Record 1: Sugar chain A, monosaccharide unit: GlcNAc, modification group: Sulfate, GU value: 5.2

[0034] Record 2: Sugar chain A, monosaccharide unit: GlcNAc, modification group: Sulfate, GU value: 5.2

[0035] Record 3: Sugar chain A, monosaccharide unit: GlcNAc, modification group: Sulfate, GU value: 5.1

[0036] Specifically, when removing redundant data, first find the exact same records through a hash algorithm or string comparison. For example, if the contents of record 1 and record 2 are exactly the same, one of the records can be deleted directly, leaving only one. For record 3, since there is a slight difference in the GU value, the record can be retained by calculating the average value or selecting the latest data. This process can be completed using the following formula: GU value of the retained record = (5.2 + 5.1) / 2 = 5.15, and the only retained record is:

[0037] Record 1: Sugar chain A, monosaccharide unit: GlcNAc, modification group: Sulfate, GU value: 5.15

[0038] (2) Filling missing values

[0039] For example, the modification group information of some glycan structures is missing, especially for glycan B:

[0040] Record 1: Sugar chain B, monosaccharide unit: Gal, modification group: missing, GU value: 3.8

[0041] Record 2: Sugar chain C, monosaccharide unit: Gal, modification group: Phosphate, GU value: 3.9

[0042] When filling in missing values, you can infer or fill in information based on sugar chains with similar structures. For example, by comparing the monosaccharide units and GU values ​​of sugar chains B and C, it is found that they have the same monosaccharide unit (Gal) and similar GU values, so it is inferred that sugar chain B may have the same modification group (Phosphate). If similar structures cannot be found, you can use mean interpolation or prediction models to fill in missing values. For example, assuming that the average GU value of other similar sugar chains in the database is 4.0, you can use mean interpolation: the GU value after filling = 4.0, and the final filled record is:

[0043] Record 1: Sugar chain B, monosaccharide unit: Gal, modification group: Phosphate, GU value: 4.0

[0044] (3) Standardized sugar chain structure information

[0045] The purpose of standardization is to unify records with different names or formats into a consistent format. Rule mapping or regular expressions can be used to achieve naming standardization. In addition, converting sugar chain structures into SMILES format is also a standardization process. For example, assuming that the structure of sugar chain D is described as: sugar chain D, monosaccharide unit: GlcNAc, modification group: Sulfate, its corresponding SMILES representation can be:

[0046]

[0047] Step S30: converting the known sugar chain structure into a known sugar chain feature vector, and using the known GU value as the target variable of the sugar chain structure model;

[0048] It should be noted that the known sugar chain feature vector refers to the numerical expression used to represent the sugar chain structure. These feature vectors contain the structural information of the sugar chain and convert it into numerical features that can be processed by the machine learning model. Specifically, the feature vector includes the following:

[0049] Monosaccharide unit: the type of each monosaccharide in the sugar chain, such as glucose (Glc), galactose (Gal), N-acetylglucosamine (GlcNAc), etc. These units can be encoded as numerical features;

[0050] Glycosidic bond type: describes the connection between monosaccharide units in the sugar chain, such as α-1,4 or β-1,6 connection. The type of these bonds is also encoded as part of the feature vector;

[0051] Modification groups: Chemical modifications on sugar chains, such as phosphate groups and sulfate groups, can affect the function and structure of sugar chains and also need to be included in the feature vector;

[0052] Spatial conformation: describes the arrangement of sugar chains in three-dimensional space, such as the configuration of chiral centers (R or S), which is very important for reflecting the stereochemical properties of sugar chains;

[0053] Molecular weight: The molecular weight of each monosaccharide unit and its modified groups can be used as part of the feature vector to reflect the quality characteristics of the sugar chain;

[0054] By converting this structural information into numerical feature vectors, machine learning models can analyze and process sugar chain structures; the target variable of the sugar chain structure model refers to the value that the model needs to predict during the training process.

[0055] It is understandable that Extended Connectivity Fingerprints (ECFP) is a molecular representation method widely used in cheminformatics, especially suitable for capturing local features of molecular structures. It generates a set of fixed-length binary vectors by traversing each atom of the molecule and its chemical environment, which can be used as feature vectors in machine learning models.

[0056] It should be understood that the generation of the feature vector includes encoding the sugar chain structure using the molecular fingerprint ECFP, and the specific implementation steps may be:

[0057] (1) Construction of molecular graph

[0058] First, the sugar chain structure is represented as a graph, in which nodes represent atoms (mainly carbon, oxygen, etc. in sugar chains) and edges represent chemical bonds (such as CO bonds, CC bonds, etc.).

[0059] For example, consider a simple sugar unit (such as glucose, Glc):

[0060] Nodes: C1, C2, C3, C4, C5, C6, O1, O2, O3, O4, O5

[0061] Edges: C1-O1, C2-O2, C3-O3, C4-O4, C5-O5, C6-O6

[0062] (2) Initialize node features

[0063] In order to start the generation of feature vectors, we first need to assign initial features to each node (atom). These features can be the type of atom (such as C, O), the type of bond connected (such as single bond, double bond), whether it is in a ring, the charge of the atom, etc.

[0064] Example: Assuming that the carbon atom is defined as "1", the oxygen atom is defined as "2", and the single bond is defined as "1", the initial eigenvector of C1 may be: [1,0], indicating a carbon atom and a single bond.

[0065] (3) Recursive expansion of adjacent atoms

[0066] The core of the ECFP algorithm is to generate a feature vector that captures the chemical environment by gradually traversing each node and its neighboring nodes. This process will be iterated multiple times to capture the local structural features in the sugar chain.

[0067] Example: In actual applications, feature expansion is performed on C1 and its directly connected O1 and C2, and the feature vector of C1 is updated to [1, 0, 2, 0], and then expanded to the neighbors of C2.

[0068] (4) Iterative calculation of eigenvectors

[0069] The process is expanded through multiple iterations to capture a wider range of chemical information. For a complex sugar chain, multiple iterations may be required to ensure that all relevant structural features are included in the feature vector.

[0070] Example: After many iterations, the final feature vector of C1 may include all the information of its second-order or third-order neighbors, such as [C1,O1,C2,O2,C3,…].

[0071] (5) Generate ECFP fingerprint of the whole molecule

[0072] After the feature vectors of all nodes are generated, we merge them and convert them into a fixed-length binary vector through a hash function to form the final molecular fingerprint. This fingerprint will be used as the input feature of the machine learning model.

[0073] Example: For glucose, assume that the ECFP fingerprint generated by the above steps is [10101011...]. This vector not only captures the local features of the sugar chain, but also reflects its global structure.

[0074] Ultimately, these feature vectors will be fed into a machine learning model to train the model along with known GU values. The goal of the model is to be able to accurately predict the GU value and possible structure of an unknown glycan based on its feature vector in the future.

[0075] Step S40: arranging the sugar chain structure feature vectors to obtain a sugar chain structure feature sequence, selecting a graph convolutional network model to process the sugar chain structure feature sequence, and training the sugar chain structure model using a cross-validation method;

[0076] It can be understood that in step S30, the sugar chain structure has been converted into feature vectors, which contain the structural information of each sugar chain node (monosaccharide units and their connection methods). However, a single feature vector can only represent the local information of a node or fragment. In order to fully represent the structure of the entire sugar chain, we need to arrange and combine these feature vectors to form an overall feature sequence. Arrange feature vectors: First, sort the feature vectors of all nodes according to the topological structure of the sugar chain (such as the order from root to end). Sorting can be based on methods such as topological sorting, breadth-first search (BFS) or depth-first search (DFS). Generate feature sequence: Concatenate the sorted feature vectors into a feature sequence. This sequence can be represented as a two-dimensional matrix, in which each row represents the feature vector of a node and the columns represent different features. For example, the feature vector of node 1 (C1) is [0.2, 0.5, 0.8], the feature vector of node 2 (C2) is [0.3, 0.6, 0.9], and the feature vector of node 3 (C3) is [0.4, 0.7, 1.0]. After arrangement, the feature sequence of the sugar chain can be expressed as:

[0077]

[0078] It should be understood that a graph convolutional network (GCN) is a neural network model that processes graph structured data and can perform convolution operations on a graph to extract high-level features of nodes and their neighborhoods. In sugar chain structure analysis, GCN can capture the complex relationships between sugar chain nodes (monosaccharide units) and generate global features for classification or regression. The specific implementation steps are as follows:

[0079] GCN performs a weighted summation of the feature vector of each node and the feature vector of its neighbors, and then generates a new feature representation through an activation function (such as ReLU). For node i, the update formula of GCN can be expressed as:

[0080]

[0081] in, is the feature representation of node i at the k+1th layer, N i is the neighbor set of node i, W (k) is the weight matrix of the kth layer, c ij is the normalization coefficient between nodes i and j, and σ is the activation function (such as ReLU);

[0082] Afterwards, the feature sequence of the sugar chain structure is taken as input, and GCN gradually extracts global features through multiple layers of convolution operations, and finally generates a global feature vector for regression (GU value prediction).

[0083] It should be understood that through cross-validation, we can effectively prevent the model from overfitting and ensure that the model has good generalization ability on unseen data. The specific implementation steps of training the model using the cross-validation method can be to divide the sugar chain data set into K subsets (usually 5 or 10), use K-1 subsets for training each time, and the remaining 1 subset is used for verification. Repeat K times, and the final model performance is evaluated by the average performance of all verification sets.

[0084] Step S50: inputting the calculated GU value of the unknown sugar chain sample into the pre-trained sugar chain structure model to perform sugar chain structure similarity analysis;

[0085] It should be noted that this model is trained based on a known sugar chain dataset and can match and compare the input sugar chain sample features with the existing sugar chain structures in the database. The model training process utilizes the feature vector generated in step S30 and the global structural features extracted by the graph convolutional network (GCN) in step S40. The trained model has the ability to predict and analyze the similarity of unknown sugar chain structures.

[0086] It is understandable that when we obtain the GU value of an unknown sugar chain sample, this value is input into the model as part of the model input. The model will use the known GU value and sugar chain structure features to perform similarity analysis and predict possible sugar chain structures.

[0087] It should be understood that the model will evaluate the similarity of these sugar chain structures by calculating the similarity between the feature vector of the input sugar chain sample and the existing sugar chain feature vector in the database. The specific similarity measurement methods used include cosine similarity, Euclidean distance, or Tanimoto coefficient, etc. Based on the similarity analysis, the model will return one or more possible sugar chain structures, showing the degree of similarity between these structures and the input sugar chain sample. The higher the similarity, the higher the match between the structure and the sample.

[0088] Based on the first embodiment, a second embodiment of the present invention is proposed. In this embodiment, step S40 further includes obtaining a constrained sugar chain structure model using L2 regularization.

[0089] It should be noted that overfitting refers to the phenomenon that the model performs well on the training set but performs poorly on new data. Overfitting usually occurs when the model learns noise or details in the training data, while these features are not prevalent in the test data. L2 regularization reduces the overfitting problem by adding a penalty term of the sum of squared weights to the loss function to prevent the model parameters from becoming too large.

[0090] It should be understood that when training the sugar chain structure prediction model, the specific implementation steps of L2 regularization can be:

[0091] (1) Set the L2 regularization coefficient

[0092] First, you need to set an initial L2 regularization coefficient λ. This coefficient can be determined empirically or through cross-validation in preliminary experiments on the model. For example, you can choose λ = 0.01 as the initial value and adjust the value according to the model performance in subsequent training.

[0093] (2) Modify the loss function and add an L2 regularization term based on the original loss function.

[0094] (3) During the training process, the L2 regularization term will impose constraints on each weight update to prevent the weight parameters from becoming too large, thereby suppressing the model from overfitting to the noise in the training data. By observing the changes in the loss of the training set and the validation set, it can be determined whether the regularization has played the expected role. If the performance of the validation set is improved and the model is no longer overfitted, it can be considered that the regularization coefficient λ is set appropriately.

[0095] (4) Adjust the regularization strength. If the model is still overfitting or underfitting after initial training (i.e., the performance on the validation set is not as expected), the regularization strength can be adjusted by increasing or decreasing the value of λ. For example, if the validation set performs poorly, λ can be gradually reduced, such as to λ = 0.005, and then retraining and validation can be performed.

[0096] (5) Determination of the final model: After multiple rounds of training and validation, an optimal λ value is determined so that the model performs best on the validation set. The final model will be used in practical applications for the identification of unknown sugar chain structures.

[0097] In addition, it is understandable that traditional L2 regularization is based on the assumption of frequentism, and the regularization terms of all parameters are applied uniformly. However, in practical applications, some parameters may have greater uncertainty, so different degrees of regularization are required.

[0098] It should be understood that in order to solve the above problems, this embodiment also introduces Bayesian inference on the basis of traditional L2 regularization, regards the weights of the model as random variables, and gives them a normal distribution prior. During the training process, the distribution of weights is adjusted through posterior inference, thereby realizing adaptive regularization: p(w i ~N(0,σ 2 ))

[0099] Maximizing the posterior distribution is equivalent to minimizing the following objective function:

[0100]

[0101] Through Bayesian inference, the model can not only obtain the optimal parameter estimates, but also obtain the confidence intervals of these parameters, thereby quantifying the uncertainty of the model. This is particularly useful in the prediction of sugar chain structures, and can help determine which structural predictions are more reliable and which require further experimental verification.

[0102] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A novel sugar chain structure identification method based on GU value, characterized in that: The method comprises the following steps: Step S10: obtaining the migration time of the unknown sugar chain sample by capillary electrophoresis experiment, and calculating the GU value of the unknown sugar chain sample in combination with the migration time of the adjacent standard sample; Step S20: collecting structural information of known sugar chains and calculating corresponding GU values, organizing the structural information and corresponding GU values ​​into a data set, and obtaining the known sugar chain structure after data cleaning, which is expressed in the form of SMILES; Step S30: converting the known sugar chain structure into a known sugar chain feature vector, and using the known GU value as the target variable of the sugar chain structure model; Step S40: arranging the sugar chain structure feature vectors to obtain a sugar chain structure feature sequence, selecting a graph convolutional network model to process the sugar chain structure feature sequence, and training the sugar chain structure model using a cross-validation method; Step S50: inputting the calculated GU value of the unknown sugar chain sample into the pre-trained sugar chain structure model to perform sugar chain structure similarity analysis.

2. The novel sugar chain structure identification method based on GU value according to claim 1, characterized in that: The step S10 includes calibrating the migration time of the standard sample using an internal standard.

3. The novel sugar chain structure identification method based on GU value according to claim 1, characterized in that: The structural information of the known sugar chain in step S20 includes glycosidic bond type, monosaccharide unit, modification group, spatial conformation and molecular weight information.

4. The novel sugar chain structure identification method based on GU value according to claim 1, characterized in that: The data cleaning step in step S20 includes removing redundant data, filling missing values, and standardizing sugar chain structure information.

5. The novel sugar chain structure identification method based on GU value according to claim 1, characterized in that: In the step S30, the generation of the feature vector includes encoding the sugar chain structure using the molecular fingerprint ECFP.

6. The novel sugar chain structure identification method based on GU value according to claim 1, characterized in that: The step S40 further includes obtaining a constrained sugar chain structure model using L2 regularization.

7. A novel sugar chain structure identification device based on GU value, characterized in that: The novel sugar chain structure identification device based on GU value comprises: A GU value calculation module is used to obtain the migration time of the unknown sugar chain sample through a capillary electrophoresis experiment, and calculate the GU value of the unknown sugar chain sample in combination with the migration time of the adjacent standard sample; A data collection and processing module is used to collect the structural information of known sugar chains and calculate the corresponding GU values, organize the structural information and the corresponding GU values ​​into a data set, and represent the known sugar chain structure in the form of SMILES after data cleaning; A feature vector generation module is used to convert the known sugar chain structure into a known sugar chain feature vector, and use the known GU value as the target variable of the sugar chain structure model; a model training module, for arranging sugar chain structure feature vectors to obtain a sugar chain structure feature sequence, selecting a graph convolutional network model to process the sequence, and training a sugar chain structure model using a cross-validation method; The similarity analysis module is used to input the calculated GU value of the unknown sugar chain sample into the pre-trained sugar chain structure model to perform sugar chain structure similarity analysis.

Images