Automatic interpretation method and system for antibody identification results based on vector space model
The automatic interpretation of antibody identification results using a vector space model solves the problems of time-consuming, labor-intensive, and inaccurate manual interpretation in existing technologies, achieving rapid, accurate, and standardized antibody identification with strong fault tolerance and pattern discovery capabilities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TIANJIN MEDICAL UNIVERSITY GENERAL HOSPITAL
- Filing Date
- 2026-01-06
- Publication Date
- 2026-06-05
AI Technical Summary
Existing antibody identification methods rely on manual interpretation, which is time-consuming, labor-intensive, inaccurate, and lacks standardization. They also struggle to handle complex situations such as mixed antibodies, dose effects, and the absence of high-frequency antigens.
An automatic interpretation method based on a vector space model is adopted. By constructing an antibody-specific vector model, the antibody identification pedigree data is mapped into a fixed-dimensional vector. Dimensionality reduction is performed using PCA, autoencoders, or word embedding models, and the similarity between the vector to be tested and the antibody-specific vector is calculated to achieve automatic interpretation.
It improves the efficiency of antibody identification, reduces the subjectivity and fatigue errors of manual interpretation, and makes the results more objective and standardized. It has strong fault tolerance and fuzzy matching capabilities and can discover potential patterns.
Smart Images
Figure CN122157786A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical testing and bioinformatics technology, and in particular relates to an automatic interpretation method and system for antibody identification results based on a vector space model. Background Technology
[0002] In transfusion medicine and immunohematology, antibody identification is a crucial step in ensuring safe transfusions. Currently, conventional antibody identification methods rely on the use of commercially available antibody identification panel cells. Panel cells contain cells with multiple known antigen profiles. Multiple panel cells are used in antibody identification experiments. The original phenomena observed after the serum sample reacts with each panel cell are taken as the reaction result for that panel cell. By treating these discrete reaction results as a whole, a reaction pattern is obtained, which is then used for analysis and comparison.
[0003] The interpretation process in the existing technology is briefly described as follows: 1. Obtaining reaction results: The experiment obtains the reaction results between the test sample and a group of cells (usually 10-16). The reaction results are expressed as positive "1" or negative "0". 2. Forming a reaction pattern: Arranging these discrete reaction results into a complete "reaction pattern" is called a reaction pattern; taking the reaction results of 11 spectroscopic cells as an example, the reaction pattern is as follows: 1 1 1 1 0 0 0 0 0 0 0.
[0004] 3. Manual comparison and interpretation: The reaction pattern is compared with the antibody identification phylogenetic table.
[0005] An antibody identification pedigree chart is a two-dimensional table. Rows represent different pedigree cells, and columns represent antibodies of different specificities (e.g., anti-D, anti-E, etc.). A value of "1" indicates that the pedigree cell contains this antigen; a value of "0" indicates that the pedigree cell does not contain this antigen. This is essentially a "1-of-NEencoding" or "One-Hot Encoding" pattern. Figure 1 The table shown is a simplified phylogenetic table, which includes the 11 cell types mentioned in step 2 and 26 antibodies, arranged in 11 rows and 26 columns, where D represents anti-D, E represents anti-E, and so on.
[0006] 4. Interpretation of Results: Based on the reaction pattern in step 2, the results are manually interpreted... Figure 1 A comparison of the pedigree chart revealed that this reaction pattern perfectly matched the 1 1 1 1 0 0 0 0 0 0 0 in the anti-D column, thus identifying it as anti-D.
[0007] The above process seems to be a simple and easy operation to compare and match using human vision. However, the actual scenario is not so simple. In actual operation, due to reasons such as mixed antibodies, dose effect, and lack of high-frequency antigens, the reaction pattern formed in the experiment cannot be completely matched with the spectrum.
[0008] Specifically, this manifests as follows: Mixed antibodies: Two or more antibodies are present in the serum sample to be tested, and the reaction pattern is the "superposition result" of multiple antibody reaction modes.
[0009] Dosage effect: Some antigens react strongly on homozygous cells and weakly or not at all on heterozygous cells, resulting in "weak positive" or "suspicious" reactions in the reaction pattern. These reactions cannot be accurately described by a simple "1 / 0" and may include intensity scores such as 1+, 2+, 3+, 4+ to indicate the strength of the reaction; ± to indicate a weak positive reaction; and H to indicate a hemolytic reaction, etc.
[0010] High-frequency antigen deficiency: Cells lacking high-frequency antigens cannot find enough negative cells to confirm antibody-specific response patterns, which may result in all spectrum cells being "1".
[0011] Other factors include additional reactions caused by unknown antibodies or non-specific reactions, dose effects or antigen expression variations leading to unresponsive expected positive cells, and false positives / false negatives caused by errors in automated instruments.
[0012] Because the reaction pattern and pedigree may not match perfectly, manual interpretation cannot accurately handle fuzzy matches, and the experience and judgment standards of different technicians may differ, resulting in a high degree of arbitrariness in the interpretation results. At the same time, fatigue and errors are more likely to occur, which is not only time-consuming and laborious, but also cannot achieve rapid, accurate and standardized antibody identification. Summary of the Invention
[0013] The purpose of this invention is to propose an automatic interpretation method and system for antibody identification results based on a vector space model. By calculating vector similarity, the antibody identification results can be automatically interpreted, enabling rapid, accurate, and standardized antibody identification.
[0014] To achieve the above objectives, the technical solution of the present invention is implemented as follows: An automatic interpretation method for antibody identification results based on a vector space model includes: S1. Construct an antibody-specific vector model, mapping the antigen spectrum corresponding to each antibody specificity in the antibody identification lineage table to a fixed-dimensional antibody-specific vector; S2. The antibody identification reaction results of the serum sample to be tested are mapped to the test vector through the antibody specificity vector model; S3. Calculate the similarity between the vector to be tested and the antibody-specific vector; S4. Based on the similarity calculation results, the antibody identification results are obtained by interpreting the results according to the interpretation logic.
[0015] Furthermore, step S1 specifically includes: S101. Convert the antibody identification pedigree data into an M×N matrix, where M represents antibody specificity and N represents the pedigree cells; S102. Each row of the matrix represents an antigen spectrum specific to an antibody, and each row of the matrix is regarded as an independent data sample. S103. Using a vectorization algorithm, each data sample is mapped to a fixed-dimensional vector space to obtain the corresponding antibody-specific vector.
[0016] Preferably, the vectorization algorithm in step S103 is principal component analysis (PCA), which reduces the dimensionality of the matrix to obtain a fixed-dimensional antibody-specific vector.
[0017] Preferably, the vectorization algorithm in step S103 is an autoencoder algorithm, which generates a fixed-dimensional antibody-specific vector for each antibody-specificity through a neural network.
[0018] Preferably, the vectorization algorithm in step S103 is a word embedding model algorithm, which treats the spectrum cells represented by each column of the matrix as a context, and the value of the antibody-specific positive reaction represented by each row as a word appearing in the context. The distributed representation of each word is learned through the word embedding model, and a fixed-dimensional antibody-specific vector is output for each word, that is, for each antibody-specificity.
[0019] Furthermore, in step S3, the similarity between the vector to be tested and the antibody-specific vector is calculated using a cosine similarity algorithm, an Euclidean distance algorithm, or a dot product algorithm.
[0020] Furthermore, the judgment logic described in step S4 includes: Set the high threshold, low threshold, and mixed threshold for similarity in descending order; When the similarity is greater than or equal to the high threshold, the corresponding antibody is taken as the antibody identification result and identified as a single antibody. When the similarity is less than the high threshold and greater than or equal to the low threshold, the corresponding antibody is likely to be present and needs further confirmation. When the similarity is less than the low threshold and greater than or equal to the mixed threshold, the identification result outputs the two or more antibodies with the highest similarity, and makes a comprehensive judgment based on the logical rules of antibody appearance. When the similarity is less than the mixing threshold, it indicates that the antibody cannot be clearly identified or that an unknown antibody exists, and it is recommended to conduct additional tests.
[0021] In another aspect, this invention proposes an automatic interpretation system for antibody identification results based on a vector space model, comprising: Model building module: Constructs an antibody-specific vector model, mapping the antigen spectrum corresponding to each antibody specificity in the antibody identification lineage table to a fixed-dimensional antibody-specific vector; The test vector module maps the antibody identification reaction results of the serum sample to be tested into a test vector through the antibody specificity vector model. Similarity calculation module: Calculates the similarity between the vector to be tested and the antibody-specific vector; Interpretation module: Based on the similarity calculation results, the module performs interpretation based on interpretation logic to obtain the antibody identification results.
[0022] Furthermore, the model building module includes: Matrix unit: Converts antibody identification pedigree data into an M×N matrix, where M represents antibody specificity and N represents pedigree cells; Sample Unit: Each row of the matrix represents an antigen spectrum corresponding to a specific antibody, and each row of the matrix is regarded as an independent data sample; Vectorization unit: Using a vectorization algorithm, each data sample is mapped to a fixed-dimensional vector space to obtain the corresponding antibody-specific vector.
[0023] Preferably, the vectorization algorithm in the vectorization unit is principal component analysis (PCA), which reduces the dimension of the matrix to obtain a fixed-dimensional antibody-specific vector. or: The vectorization algorithm described in the vectorization unit is an autoencoder algorithm, which generates a fixed-dimensional antibody-specific vector for each antibody-specificity through a neural network; or: The vectorization algorithm described in the vectorization unit is a word embedding model algorithm. It treats the spectral cells represented by each column of the matrix as a context, and the value of the antibody-specific positive reaction represented by each row as a word appearing in that context. It learns the distributed representation of each word through the word embedding model, and outputs a fixed-dimensional antibody-specific vector for each word, that is, for each antibody specificity.
[0024] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention maps antibody identification results into vectors that can be used for similarity calculation through a vector space model. By calculating similarity, the automatic interpretation of antibody identification results is realized, which significantly improves work efficiency and frees up manpower.
[0025] 2. This invention avoids the subjectivity and fatigue errors of manual interpretation, resulting in more objective and standardized results.
[0026] 3. This invention has strong fault tolerance and fuzzy matching capabilities. Vector similarity calculation can effectively handle the impact of "extra responses" or "missing responses" on the response pattern and find the most likely match.
[0027] 4. This invention can discover potential patterns. In vector space, antibodies with similar antigenicity (such as those belonging to the same blood group system) may have similar vector positions, which helps to discover potential patterns that are difficult to detect manually.
[0028] 5. This invention is easy to integrate and upgrade. The proposed method and system can be easily integrated into existing laboratory information systems, and when there are new spectra of cell or antibody specificity, it can be updated simply by readjusting the vector space model. Attached Figure Description
[0029] Figure 1 It is a simplified genealogical table.
[0030] Figure 2 This is a flowchart of Embodiment 1 of the present invention.
[0031] Figure 3 This is a schematic diagram illustrating an application example of Embodiment 2 of the present invention.
[0032] Figure 4 This is a schematic diagram of the system structure of Embodiment 3 of the present invention. Detailed Implementation
[0033] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0034] The present invention will be further described below with reference to specific embodiments and accompanying drawings.
[0035] Example 1: In this embodiment, the automatic interpretation method for antibody identification results based on the vector space model is as follows: Figure 2 As shown, it includes the following steps: Step 1: Constructing an antibody-specific vector model (training phase): 1. Obtain antibody identification pedigree data and convert it into an M×N matrix, where M represents antibody specificity (e.g., anti-D, anti-C, anti-E, etc.) and N represents the pedigree cells.
[0036] 2. Treat each row of the matrix (each row of the matrix corresponds to an antibody-specific antigen spectrum) as an independent data sample.
[0037] At this point, we can directly use the rows of the matrix as high-dimensional vectors and then perform similarity calculations through normalization and other processing. However, this method is computationally complex, and the computational efficiency and accuracy of the results will be affected. It may even fail to accurately reflect the true similarity. Therefore, we must proceed to the next step.
[0038] 3. Using a vectorization algorithm, each antibody-specific data sample is mapped to a D-dimensional vector space to obtain an antibody-specific vector. The vectorization algorithm includes, but is not limited to: Principal Component Analysis (PCA): Reduces the dimensionality of the phylogenetic table matrix and extracts the main characteristic components.
[0039] Autoencoder: Learns an efficient, low-dimensional representation of a genealogy table through a neural network.
[0040] Word embedding model (such as the CBOW / Skip-gram idea of Word2Vec): Each spectrum cell is regarded as a "context", and the occurrence of a positive reaction (1) is regarded as a "word" (i.e. antibody specificity) appearing in that context.
[0041] 3.1 Principal Component Analysis (PCA): Dimensionality reduction using PCA may include the following steps: (1) Matrix preprocessing: The antibody identification pedigree data used for training were all converted into an M×N matrix. The matrix data were then mean-centered and normalized to obtain the preprocessed matrix.
[0042] (2) Calculate the covariance matrix: The covariance matrix of the preprocessed matrix is calculated, and the linear correlation between different features (spectral cells) is measured by the covariance matrix.
[0043] (3) Solution and arrangement: Find the eigenvalues and corresponding eigenvectors of the covariance matrix; Arrange all eigenvalues in descending order. The corresponding eigenvectors are also rearranged accordingly.
[0044] (4) Construct the projection matrix: Calculate the proportion of each eigenvalue to the sum of all eigenvalues, as the contribution rate; Set a contribution rate threshold and calculate the minimum number of D feature values whose cumulative contribution rate exceeds the threshold. Select the eigenvectors corresponding to the top D largest eigenvalues as columns, and form an N×D matrix as the projection matrix.
[0045] (5) Obtain the antibody specificity vector: Multiplying the preprocessed matrix by the projection matrix yields an M×D matrix, where each row of the matrix is the corresponding antibody-specific vector.
[0046] The projection matrix, as a vector space model, can be used to vectorize the reaction results of antibody identification in the serum sample to be tested.
[0047] 3.2 Self-encoder: (1) Matrix preprocessing: The antibody identification pedigree data used for training were all converted into an M×N matrix, and the matrix data were normalized to obtain the preprocessed matrix.
[0048] (2) Constructing an autoencoder model: The design includes a symmetrical neural network structure with encoder and decoder; The encoder gradually reduces the dimensionality, using multiple hidden layers to reach the required D dimension; the decoder, on the other hand, gradually increases the dimensionality to N dimensions using multiple hidden layers.
[0049] (3) Training: The Adam optimization algorithm is selected, with mean squared error or binary cross-entropy as the loss function. The preprocessed matrix is used as both the input and target output of the model for training.
[0050] During training, the weights in the network are continuously adjusted to minimize the reconstruction error, so that the decoder output is as close as possible to the original input.
[0051] (4) Obtain the antibody specificity vector: After training, the encoder model is used alone as the vector space model. The matrix transformed from the antibody identification pedigree data used for interpretation is input into the vector space model to obtain the antibody specificity vector.
[0052] The vector space model can be used to vectorize the reaction results of the next step of antibody identification of the serum sample to be tested.
[0053] 3.3 Word Embedding Model: Each column of the matrix represents a spectrum of cells as a context, and each row represents a value of antibody-specific positive reaction as a word appearing in that context. The distributed representation of each word is learned through a word embedding model, and a fixed-dimensional antibody-specific vector is output for each word, that is, for each antibody specificity.
[0054] Specifically, it includes: (1) Convert the M×N matrix data into a context list: The antibody identification pedigree data used for training are all transformed into an M×N matrix. Each column represents a spectral cell as a context, and all antibodies that produce a positive reaction on this spectral cell are regarded as words in this context. N contexts are constructed to form a corpus, which serves as the raw data for the input word embedding model.
[0055] (2) Training: You can choose a commonly used word embedding model, such as Word2Vec or GloVe.
[0056] For each sentence in the input corpus, Word2Vec trains a neural network and optimizes the prediction task by adjusting the weights to obtain a vector space model.
[0057] GloVe constructs an M×M co-occurrence matrix, where each element represents the number of times the antibody in its row co-occurs with the antibody in its column. GloVe then learns two vectors corresponding to each antibody, setting the objective function as follows: the dot product of the two vectors should approximately equal the logarithm of their co-occurrence counts. By minimizing this objective function, GloVe decomposes the co-occurrence matrix to obtain the specific vector for each antibody. This trained vector space model is then used to vectorize the reaction results for antibody identification in the next step of testing serum samples.
[0058] Step 2: Vectorize the results of the test specimen (application stage): 1. Obtain the reaction results of the serum sample to be tested with N spectrophotometers and form a 1×N binary vector (e.g., [1,0, 1, 1, 0, ...]).
[0059] 2. The reaction results of antibody identification of the serum specimens to be tested are vectorized using the aforementioned vector space model.
[0060] The same vector space model used in step one to generate the antibody-specific vector will be used in this step to input the binary vector to obtain the vector to be tested.
[0061] Step 3: Perform similarity calculation: 1. Calculate the similarity between the vector to be tested and all M antibody-specific vectors obtained in step one.
[0062] 2. The similarity calculation methods include, but are not limited to: cosine similarity, Euclidean distance, and dot product.
[0063] Cosine similarity is preferred because it focuses more on the direction of vectors than their magnitude, and is better able to resist the influence of fluctuations in reaction intensity.
[0064] Step 4: Based on the similarity calculation results, perform interpretation based on the interpretation logic to obtain the antibody identification results.
[0065] The similarity scores are sorted according to the similarity calculation results. For example: Anti-D: similarity 0.98; Anti-E: similarity 0.65; Anti-M: similarity 0.55; ...
[0066] The interpretation logic includes: Set the high threshold, low threshold, and mixed threshold for similarity in descending order; When the similarity is greater than or equal to the high threshold, the corresponding antibody is taken as the antibody identification result and identified as a single antibody. When the similarity is less than the high threshold and greater than or equal to the low threshold, the corresponding antibody is likely to be present and needs further confirmation. When the similarity is less than the low threshold and greater than or equal to the mixed threshold, the identification result outputs the two or more antibodies with the highest similarity, and makes a comprehensive judgment based on the logical rules of antibody appearance (such as antigen pairs: anti-c and anti-E often appear at the same time); When the similarity is less than the mixing threshold, it indicates that the antibody cannot be clearly identified or that an unknown antibody exists, and it is recommended to conduct additional tests.
[0067] The automatic interpretation method for antibody identification results based on the vector space model proposed in this embodiment maps the antibody identification results into vectors that can be used for similarity calculation. Through similarity calculation, the automatic interpretation of antibody identification results is realized, which significantly improves work efficiency, frees up manpower, and makes the results more objective and standardized. It also has strong fault tolerance and fuzzy matching capabilities.
[0068] Example 2: This embodiment is an application example of the method described in Embodiment 1, such as... Figure 3 As shown: 1. Training Phase: Suppose there is an antibody identification pedigree table containing 5 antibodies (anti-A, anti-B, anti-D, anti-C, anti-E) and 10 spectral cells.
[0069] Transform it into a 5x10 matrix, and then vectorize this matrix. The process involves training a vector space model, which can be done using algorithms such as PCA or Autoencode.
[0070] This embodiment reduces the specificity of each antibody from 10 dimensions to 2 dimensions by using a vector space model based on the PCA algorithm.
[0071] get: Anti-A vector: [0.12, 0.85]; Anti-B vector: [0.92, -0.15]; Anti-D vector: [0.45, 0.55];
[0072] 2. Application Phase: The reaction result of obtaining a serum sample to be tested constitutes the reaction pattern, namely [0, 1, 1, 0, 1, 0, 1, 0, 1, 1] (10 dimensions).
[0073] The same vector space model using the PCA algorithm is used for vectorization, and it is projected into a 2D space to obtain the test vector [0.50, 0.60].
[0074] 3. Vector similarity calculation: Calculate the cosine similarity between the test vector [0.50, 0.60] and each antibody vector: Sim(tested, anti-A) = 0.75; Sim(to be tested, anti-B) = 0.25; Sim(to be tested, anti-D) = 0.98;
[0075] 4. Result Interpretation: In this embodiment, the high threshold is set to 0.9, the low threshold to 0.7, and the mixed threshold to 0.15.
[0076] Based on the interpretation logic, since the similarity with anti-D (0.98) is the highest and much higher than others, the system automatically outputs the identification result as "anti-D" and gives a high confidence level, confirming it as a single antibody.
[0077] Example 3: This embodiment proposes an automatic interpretation system for antibody identification results based on a vector space model, such as... Figure 4 As shown, it includes: Model building module: Constructs an antibody-specific vector model, mapping the antigen spectrum corresponding to each antibody specificity in the antibody identification lineage table to a fixed-dimensional antibody-specific vector; The test vector module maps the antibody identification reaction results of the serum sample to be tested into a test vector through the antibody specificity vector model. Similarity calculation module: Calculates the similarity between the vector to be tested and the antibody-specific vector; Interpretation module: Based on the similarity calculation results, the module performs interpretation based on interpretation logic to obtain the antibody identification results.
[0078] Furthermore, the model building module includes: Matrix unit: Converts antibody identification pedigree data into an M×N matrix, where M represents antibody specificity and N represents pedigree cells; Sample Unit: Each row of the matrix represents an antigen spectrum corresponding to a specific antibody, and each row of the matrix is regarded as an independent data sample; Vectorization unit: Using a vectorization algorithm, each data sample is mapped to a fixed-dimensional vector space to obtain the corresponding antibody-specific vector.
[0079] Preferably, the vectorization algorithm in the vectorization unit is principal component analysis (PCA), which reduces the dimension of the matrix to obtain a fixed-dimensional antibody-specific vector. or: The vectorization algorithm described in the vectorization unit is an autoencoder algorithm, which generates a fixed-dimensional antibody-specific vector for each antibody-specificity through a neural network; or: The vectorization algorithm described in the vectorization unit is a word embedding model algorithm. It treats the spectrum cells represented by each column of the matrix as a context, and the value of the antibody-specific positive reaction represented by each row as a word appearing in the context. It learns the distributed representation of each word through the word embedding model, and outputs a fixed-dimensional antibody-specific vector for each word, that is, for each antibody specificity.
[0080] The similarity calculation module calculates the similarity between the vector to be tested and the antibody-specific vector using a cosine similarity algorithm, an Euclidean distance algorithm, or a dot product algorithm.
[0081] The judgment logic described in the judgment module includes: Set the high threshold, low threshold, and mixed threshold for similarity in descending order; When the similarity is greater than or equal to the high threshold, the corresponding antibody is taken as the antibody identification result and identified as a single antibody. When the similarity is less than the high threshold and greater than or equal to the low threshold, the corresponding antibody is likely to be present and needs further confirmation. When the similarity is less than the low threshold and greater than or equal to the mixed threshold, the identification result outputs the two or more antibodies with the highest similarity, and makes a comprehensive judgment based on the logical rules of antibody appearance. When the similarity is less than the mixing threshold, it indicates that the antibody cannot be clearly identified or that an unknown antibody exists, and it is recommended to conduct additional tests.
[0082] The automatic antibody identification result interpretation system based on the vector space model proposed in this embodiment can realize the automatic antibody identification result interpretation method based on the vector space model described in Embodiment 1, and has the same technical effect as Embodiment 1.
[0083] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. An automatic interpretation method for antibody identification results based on a vector space model, characterized in that, include: S1. Construct an antibody-specific vector model, mapping the antigen spectrum corresponding to each antibody specificity in the antibody identification lineage table to a fixed-dimensional antibody-specific vector; S2. The antibody identification reaction results of the serum sample to be tested are mapped to the test vector through the antibody specificity vector model; S3. Calculate the similarity between the vector to be tested and the antibody-specific vector; S4. Based on the similarity calculation results, the antibody identification results are obtained by interpreting the results according to the interpretation logic.
2. The automatic interpretation method for antibody identification results based on a vector space model according to claim 1, characterized in that, Step S1 specifically includes: S101. Convert the antibody identification pedigree data into an M×N matrix, where M represents antibody specificity and N represents the pedigree cells; S102. Each row of the matrix represents an antigen spectrum specific to an antibody, and each row of the matrix is regarded as an independent data sample. S103. Using a vectorization algorithm, each data sample is mapped to a fixed-dimensional vector space to obtain the corresponding antibody-specific vector.
3. The automatic interpretation method for antibody identification results based on a vector space model according to claim 2, characterized in that, The vectorization algorithm described in step S103 is principal component analysis (PCA). PCA is used to reduce the dimensionality of the matrix to obtain a fixed-dimensional antibody-specific vector.
4. The automatic interpretation method for antibody identification results based on a vector space model according to claim 2, characterized in that, The vectorization algorithm described in step S103 is an autoencoder algorithm, which generates a fixed-dimensional antibody-specific vector for each antibody-specific antibody through a neural network.
5. The automatic interpretation method for antibody identification results based on a vector space model according to claim 2, characterized in that, The vectorization algorithm described in step S103 is a word embedding model algorithm. It treats the spectrum cells represented by each column of the matrix as a context, and the value of the antibody-specific positive reaction represented by each row as a word appearing in the context. It learns the distributed representation of each word through the word embedding model, and outputs a fixed-dimensional antibody-specific vector for each word, that is, for each antibody specificity.
6. The automatic interpretation method for antibody identification results based on a vector space model according to claim 1, characterized in that, In step S3, the similarity between the vector to be tested and the antibody-specific vector is calculated using a cosine similarity algorithm, an Euclidean distance algorithm, or a dot product algorithm.
7. The automatic interpretation method for antibody identification results based on a vector space model according to claim 1, characterized in that, The judgment logic described in step S4 includes: Set the high threshold, low threshold, and mixed threshold for similarity in descending order; When the similarity is greater than or equal to the high threshold, the corresponding antibody is taken as the antibody identification result and identified as a single antibody. When the similarity is less than the high threshold and greater than or equal to the low threshold, the corresponding antibody is likely to be present and needs further confirmation. When the similarity is less than the low threshold and greater than or equal to the mixed threshold, the identification result outputs the two or more antibodies with the highest similarity, and makes a comprehensive judgment based on the logical rules of antibody appearance. When the similarity is less than the mixing threshold, it indicates that the antibody cannot be clearly identified or that an unknown antibody exists, and it is recommended to conduct additional tests.
8. An automatic interpretation system for antibody identification results based on a vector space model, characterized in that, include: Model building module: Constructs an antibody-specific vector model, mapping the antigen spectrum corresponding to each antibody specificity in the antibody identification lineage table to a fixed-dimensional antibody-specific vector; The test vector module maps the antibody identification reaction results of the serum sample to be tested into a test vector through the antibody specificity vector model. Similarity calculation module: Calculates the similarity between the vector to be tested and the antibody-specific vector; Interpretation module: Based on the similarity calculation results, the module performs interpretation based on interpretation logic to obtain the antibody identification results.
9. The automatic interpretation system for antibody identification results based on a vector space model according to claim 8, characterized in that, The model building module includes: Matrix unit: Converts antibody identification pedigree data into an M×N matrix, where M represents antibody specificity and N represents pedigree cells; Sample Unit: Each row of the matrix represents an antigen spectrum specific to an antibody, and each row of the matrix is regarded as an independent data sample; Vectorization unit: Using a vectorization algorithm, each data sample is mapped to a fixed-dimensional vector space to obtain the corresponding antibody-specific vector.
10. The automatic interpretation system for antibody identification results based on a vector space model according to claim 9, characterized in that, The vectorization algorithm in the vectorization unit is principal component analysis (PCA), which reduces the dimension of the matrix to obtain a fixed-dimensional antibody-specific vector. or: The vectorization algorithm described in the vectorization unit is an autoencoder algorithm, which generates a fixed-dimensional antibody-specific vector for each antibody-specificity through a neural network; or: The vectorization algorithm described in the vectorization unit is a word embedding model algorithm. It treats the spectral cells represented by each column of the matrix as a context, and the value of the antibody-specific positive reaction represented by each row as a word appearing in that context. It learns the distributed representation of each word through the word embedding model, and outputs a fixed-dimensional antibody-specific vector for each word, that is, for each antibody specificity.