Prediction method, device and program product for intermediate mediterranean anemia
By extracting mass spectrometry features using blood tandem mass spectrometry technology and machine learning models or graph neural networks, the accuracy and cost issues of intermediate thalassemia screening methods have been solved, achieving efficient and reliable early prediction.
Patent Information
- Application Number
- CN202511591253.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-03
- Publication Date
- 2026-01-30
AI Technical Summary
Existing screening methods for intermediate thalassemia lack sufficient sensitivity and specificity, have a high false negative rate, and require expensive genetic testing equipment, making them difficult to apply on a large scale in clinical practice.
Mass spectrometry features such as butyrylcarnitine, octadecenoylcarnitine, glutaryl-CoA, and 3-hydroxyhexanoyl-CoA are extracted using blood tandem mass spectrometry. These features are then combined with machine learning models or graph neural networks to predict intermediate thalassemia.
It improves the accuracy and sensitivity of predicting intermediate thalassemia, reduces testing costs, and provides reliability and credibility for early prediction.
Smart Images
Figure CN121439259A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent healthcare, specifically to a method, device, program product, and computer-readable storage medium for predicting intermediate thalassemia. Background Technology
[0002] Thalassemia is one of the most common single-gene inherited diseases worldwide. Data shows that approximately 1.5% of the world's population carries the β-thalassemia allele, and 5% carries the α-thalassemia allele. Prenatal diagnosis can prevent the birth of individuals with severe thalassemia, but individuals with intermediate thalassemia may experience moderate anemia, growth retardation, skeletal abnormalities, and hepatosplenomegaly, and may require blood transfusions during stressful events such as infection or pregnancy. Currently, routinely used thalassemia screening methods, such as complete blood count, blood smears, osmotic fragility tests, and hemoglobin analysis methods (including high-performance liquid chromatography and capillary electrophoresis), have insufficient sensitivity and specificity, and a high false negative rate. Therefore, there is an urgent need to develop a new and efficient method for screening intermediate thalassemia. Existing screening methods include mass spectrometry and mathematical modeling of erythrocyte indices; however, these techniques have not yet been widely used in clinical practice due to low accuracy and the limited application of cutoff values. Genetic testing is the final step in the screening process and is the gold standard for diagnosing thalassemia. However, it requires technical support and the equipment is expensive, which increases the financial burden on ordinary families. Summary of the Invention
[0003] To address the above problems, this invention provides a method for predicting intermediate thalassemia, specifically including:
[0004] S1. Obtain the blood tandem mass spectrometry data of the subject;
[0005] S2. Based on the blood tandem mass spectrometry data, feature extraction is performed to obtain mass spectrometry features, which include: butyrylcarnitine, octadecenoylcarnitine, glutaryl-CoA, and 3-hydroxyhexanoyl-CoA.
[0006] S3. Based on the mass spectrometry characteristics, the intermediate thalassemia is predicted to obtain the prediction results.
[0007] Optionally, the mass spectrometry features may further include any one or more of the following: the ratio of free carnitine to the sum of palmitoylcarnitine and octadecylcarnitine, propionylcarnitine / acetylcarnitine, 3-hydroxyoctadecylcarnitine, isovalerylcarnitine / acetylcarnitine, 3-hydroxypalmitoylcarnitine, leucine / phenylalanine, myristoylcarnitine / palmitoylcarnitine, capryloylcarnitine / decylcarnitine, propionylcarnitine, isovalerylcarnitine, and myristoylcarnitine.
[0008] Optionally, the prediction is performed by a prediction model, and the mass spectrometry features are input into the prediction model to predict intermediate thalassemia and obtain the prediction result.
[0009] Optionally, the prediction model includes any one or more of the following: GBM, SVM, RF, GLM, KNN, LDA, decision tree, random forest, and Naive Bayes.
[0010] Optionally, the prediction model construction and training process includes:
[0011] Obtain the blood tandem mass spectrometry dataset;
[0012] After variable identification and removal from the blood tandem mass spectrometry dataset, the filtered variables are obtained.
[0013] Collinearity calculations are performed on the screened variables, and collinear variables are removed to obtain the second screened variables;
[0014] The variables after the second screening are fed into the machine learning model to be trained for feature learning to obtain the prediction model;
[0015] Optionally, the variable identification includes zero-variable variables, near-zero-variable variables, and highly correlated variables.
[0016] Optionally, S2 is replaced by: inputting the blood tandem mass spectrometry data into a second prediction model to predict intermediate thalassemia and obtain the prediction result, wherein the second prediction model is obtained through neural network training; and removing S3.
[0017] Optionally, the neural network includes any one or more of the following: one-dimensional convolutional neural network, Transformer, graph neural network.
[0018] Optionally, the neural network is a graph neural network, which is constructed by using blood cascade graph data and then trained using the graph data and domain knowledge;
[0019] Optionally, the training process of the graph neural network includes:
[0020] Obtain the blood tandem mass spectrometry dataset;
[0021] Based on the blood tandem mass spectrometry dataset, each peak detected in the mass spectrometry data is taken as a node, and edges are constructed through the biochemical relationships between the nodes to obtain mass spectrometry data.
[0022] The mass spectrum data is input into the graph neural network to be trained to obtain the graph neural network.
[0023] Optionally, the biochemical relationships include any one or more of the following: isotope relationships, charge relationships, adduct relationships, and edges are constructed through isotope relationships and / or charge relationships and / or adduct relationships between nodes.
[0024] Optionally, the biochemical relationship can be replaced with a hemoglobin relationship, with edges constructed through the hemoglobin relationships between nodes.
[0025] Optionally, the hemoglobin relationship includes different characteristic peaks of the same hemoglobin subunit, hemoglobin peptide mass, and / or hemoglobin variant peaks.
[0026] The hemoglobin variant peak is a hemoglobin variant peak known to be associated with thalassemia.
[0027] Optionally, the construction edge also includes hemoglobin relationships, which are constructed between nodes through biochemical relationships and hemoglobin relationships.
[0028] Optionally, the nodes in the mass spectrum data also include acylcarnitine. The mass spectrum data includes a node feature matrix and an adjacency matrix. Acylcarnitine features are added to the node feature matrix constructed from the nodes to obtain a second node feature matrix. The second node feature matrix and the adjacency matrix constructed from the edges form the second mass spectrum data. The second mass spectrum data is input to the graph neural network to be trained for training to obtain the graph neural network.
[0029] Optionally, the acylcarnitine includes any one or more of the following: butyrylcarnitine, octadecylcarnitine, glutaryl-CoA / 3-hydroxyhexanoyl-CoA, free carnitine to palmitoylcarnitine, the ratio of the sum of octadecylcarnitine, propionylcarnitine / acetylcarnitine, 3-hydroxyoctadecylcarnitine, isovalerylcarnitine / acetylcarnitine, 3-hydroxypalmitoylcarnitine, leucine / phenylalanine, myristoylcarnitine / palmitoylcarnitine, capryloylcarnitine / capryloylcarnitine, propionylcarnitine, isovalerylcarnitine, and myristoylcarnitine.
[0030] The purpose of this invention is to provide a computer program product comprising a computer program or instructions, wherein the computer program or instructions are executed by a processor to implement the above-described method for predicting intermediate thalassemia.
[0031] The purpose of this invention is to provide a computer device comprising a memory, a processor, and a computer program or instructions stored in the memory, wherein the computer program or instructions are executed by the processor to implement the above-described method for predicting intermediate thalassemia.
[0032] The purpose of this invention is to provide a computer-readable storage medium having a computer program or instructions stored thereon, which are executed by a processor to implement the above-described method for predicting intermediate thalassemia.
[0033] Advantages of this invention:
[0034] 1. To address the issues of insufficient sensitivity and specificity and high false negative rates in early prediction methods for neonatal intermediate thalassemia, including complete blood count, blood smear, osmotic fragility test, and hemoglobin analysis, as well as the low accuracy of mathematical modeling using mass spectrometry and erythrocyte index, this invention proposes using blood tandem mass spectrometry technology for mass spectrometry feature extraction and using these features to predict neonatal intermediate thalassemia. Among these features, butyrylcarnitine has the best prediction accuracy, followed by octadecenoylcarnitine, glutaryl-CoA, or 3-hydroxyhexanoyl-CoA. In addition, the ratio of free carnitine to palmitoyl carnitine, octadecyl carnitine, propionyl carnitine / acetyl carnitine, 3-hydroxyoctadecyl carnitine, isovaleryl carnitine / acetyl carnitine, 3-hydroxypalmitoyl carnitine, leucine / phenylalanine, myristenoyl carnitine / palmitoyl carnitine, capryloyl carnitine / decyl carnitine, propionyl carnitine, isovaleryl carnitine, and myristoyl carnitine have a good predictive effect on intermediate thalassemia and are helpful for early prediction of intermediate thalassemia in clinical practice.
[0035] 2. To address the issue of insufficient accuracy in current intermediate thalassemia detection technologies, this invention also trains a graph convolutional neural network for prediction, incorporates thalassemia-related knowledge, constructs mass spectrometry data, and utilizes the graph convolutional network to extract important features from the mass spectrometry data, thereby completing the training of the graph convolutional neural network to improve the prediction accuracy of intermediate thalassemia.
[0036] 3. To improve prediction accuracy, acylcarnitine features are added to the graph node construction when building mass spectrometry data. When constructing edges, adjacency matrices are generated using knowledge related to thalassemia. This provides richer data dimensions and related knowledge for intermediate thalassemia, improving prediction sensitivity and accuracy, while also enhancing reliability and credibility. Attached Figure Description
[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0038] Figure 1 A schematic diagram of the prediction method for intermediate thalassemia provided in an embodiment of the present invention;
[0039] Figure 2 A schematic diagram of a prediction system for intermediate thalassemia provided in an embodiment of the present invention;
[0040] Figure 3 A schematic diagram of a predictive device for intermediate thalassemia provided in an embodiment of the present invention;
[0041] Figure 4 ROC curves of indicators showing significant differences in single-factor logical analysis provided in this embodiment of the invention;
[0042] Figure 5 Calibration curves for each machine learning model provided in the embodiments of the present invention;
[0043] Figure 6 The improvement curves of various machine learning models provided in the embodiments of the present invention;
[0044] Figure 7 Decision curves for various machine learning models provided in embodiments of the present invention;
[0045] Figure 8 The relative weight values of the indicators included in each machine learning model provided in the embodiments of the present invention. Detailed Implementation
[0046] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.
[0047] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as S101, S102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.
[0048] Figure 1 A schematic diagram of the prediction method for intermediate thalassemia provided in this embodiment of the invention specifically includes:
[0049] S1. Obtain the blood tandem mass spectrometry data of the subject;
[0050] In one embodiment, two or more mass spectrometers are connected together, called tandem mass spectrometry. The simplest tandem mass spectrometry (MS / MS) consists of two mass spectrometers connected in series, where the first mass analyzer (MS1) pre-separates or adds energy to the ions, and the results are analyzed by the second-stage mass analyzer (MS2). The most common tandem mass spectrometry is triple quadrupole tandem mass spectrometry. The first and third-stage quadrupole analyzers are MS1 and MS2, respectively. The function of the second-stage quadrupole analyzer is to bombard the peaks obtained from MS1, fragmenting the parent ions before they enter MS2 for further analysis.
[0051] In one specific embodiment, multiple studies have shown that the metabolomics characteristics of thalassemia patients are significantly different from those of healthy individuals. This difference is evident in chorionic villus samples from heterozygous and homozygous β-thalassemia individuals, amniotic fluid samples from fetuses with hydrops Bartholin's disease, blood samples from β-thalassemia patients, urine samples from β-thalassemia patients receiving hydroxyurea treatment, and urine samples from patients with severe β-thalassemia. This study aims to identify specific predictive indicators for intermediate thalassemia using univariate logistic regression analysis, and simultaneously establish an intermediate thalassemia risk prediction model by analyzing neonatal blood tandem mass spectrometry data using multi-algorithm data modeling techniques. Currently, there are no studies using blood tandem mass spectrometry data to predict intermediate thalassemia; this method offers advantages such as speed, simplicity, multiple benefits, and cost-effectiveness.
[0052] S2. Based on the blood tandem mass spectrometry data, feature extraction is performed to obtain mass spectrometry features; the mass spectrometry features include: butyrylcarnitine, octadecenoylcarnitine, glutaryl-CoA, and 3-hydroxyhexanoyl-CoA.
[0053] In one embodiment, the mass spectrometry features further include any one or more of the following: the ratio of free carnitine to the sum of palmitoylcarnitine and octadecylcarnitine, propionylcarnitine / acetylcarnitine, 3-hydroxyoctadecylcarnitine, isovalerylcarnitine / acetylcarnitine, 3-hydroxypalmitoylcarnitine, leucine / phenylalanine, myristoylcarnitine / palmitoylcarnitine, capryloylcarnitine / decylcarnitine, propionylcarnitine, isovalerylcarnitine, and myristoylcarnitine.
[0054] S3. Based on the mass spectrometry characteristics, the intermediate thalassemia is predicted to obtain the prediction results.
[0055] In one embodiment, the prediction is performed by a prediction model, whereby the mass spectrometry features are input into the prediction model to predict intermediate thalassemia and obtain the prediction result.
[0056] In one embodiment, the prediction model includes any one or more of the following: GBM, SVM, RF, GLM, KNN, LDA, decision tree, random forest, and Naive Bayes.
[0057] In one embodiment, the prediction model construction and training process includes:
[0058] Obtain the blood tandem mass spectrometry dataset;
[0059] After variable identification and removal from the blood tandem mass spectrometry dataset, the filtered variables are obtained.
[0060] Collinearity calculations are performed on the screened variables, and collinear variables are removed to obtain the second screened variables;
[0061] The variables after the second screening are fed into the machine learning model to be trained for feature learning to obtain the prediction model.
[0062] Optionally, the variable identification includes zero-variable variables, near-zero-variable variables, and highly correlated variables.
[0063] In one embodiment, S2 is replaced by: inputting blood tandem mass spectrometry data into a first prediction model to predict intermediate thalassemia and obtain prediction results, wherein the first prediction model is a trained machine learning model, and the machine learning model includes any one or more of the following: GBM, SVM, RF, GLM, KNN, LDA, decision tree, random forest, Naive Bayes; and S3 is removed.
[0064] In one embodiment, S2 is replaced by: inputting the blood tandem mass spectrometry data into a second prediction model to predict intermediate thalassemia and obtain the prediction result, wherein the second prediction model is obtained through neural network training; and removing S3.
[0065] Optionally, the neural network includes any one or more of the following: one-dimensional convolutional neural network, Transformer, graph neural network.
[0066] In one embodiment, the neural network is a graph neural network, which is constructed by building graph data from blood cascade graph data and then trained using the graph data and domain knowledge.
[0067] In one embodiment, the training process of the graph neural network includes:
[0068] Obtain the blood tandem mass spectrometry dataset;
[0069] Based on the blood tandem mass spectrometry dataset, each peak detected in the mass spectrometry data is taken as a node, and edges are constructed through the biochemical relationships between the nodes to obtain mass spectrometry data.
[0070] The mass spectrum data is input into the graph neural network to be trained to obtain the graph neural network.
[0071] In one embodiment, the biochemical relationship includes any one or more of the following: isotope relationship, charge relationship, adduct relationship, and edges are constructed through isotope relationship and / or charge relationship and / or adduct relationship between nodes.
[0072] In one embodiment, the biochemical relationship is replaced by a hemoglobin relationship, with edges constructed through the hemoglobin relationships between nodes;
[0073] Optionally, the hemoglobin relationship includes different characteristic peaks of the same hemoglobin subunit, hemoglobin peptide mass, and / or hemoglobin variant peaks;
[0074] The hemoglobin variant peak is a hemoglobin variant peak known to be associated with thalassemia.
[0075] In one embodiment, the construction edge also includes hemoglobin relationships, which are constructed between nodes through biochemical relationships and hemoglobin relationships.
[0076] In one embodiment, the nodes in the mass spectrum data further include acylcarnitine. The mass spectrum data includes a node feature matrix and an adjacency matrix. Acylcarnitine features are added to the node feature matrix constructed from the nodes to obtain a second node feature matrix. The second node feature matrix and the adjacency matrix constructed from the edges form the second mass spectrum data. The second mass spectrum data is input to a graph neural network to be trained to obtain a graph neural network.
[0077] Optionally, the acylcarnitine includes any one or more of the following: butyrylcarnitine, octadecylcarnitine, glutaryl-CoA / 3-hydroxyhexanoyl-CoA, free carnitine to palmitoylcarnitine, the ratio of the sum of octadecylcarnitine, propionylcarnitine / acetylcarnitine, 3-hydroxyoctadecylcarnitine, isovalerylcarnitine / acetylcarnitine, 3-hydroxypalmitoylcarnitine, leucine / phenylalanine, myristoylcarnitine / palmitoylcarnitine, capryloylcarnitine / capryloylcarnitine, propionylcarnitine, isovalerylcarnitine, and myristoylcarnitine.
[0078] In one specific embodiment, the purpose of this invention is to introduce a quantitative method based on blood tandem mass spectrometry for rapidly predicting the risk of intermediate thalassemia. On one hand, through univariate logistic regression analysis, receiver operating characteristic (ROC) curves are analyzed to statistically identify blood tandem mass spectrometry indicators with significant differences. On the other hand, a machine learning model is constructed to distinguish between healthy controls and patients with intermediate thalassemia using blood tandem mass spectrometry data. The aim is to provide a rapid and reliable method for predicting the risk of intermediate thalassemia.
[0079] I. Introduction to Single-Factor Logistic Regression Analysis;
[0080] To study the relationship between an independent variable (explanatory variable) and a binary dependent variable (explained variable), maximum likelihood estimation (MLE) was used to estimate the model parameters β0 and β1. Blood tandem mass spectrometry indicators with significant differences between the two groups were screened out, and then ROC curve analysis was performed.
[0081] II. Introduction to Model Analysis;
[0082] In this invention, nine machine learning models were selected: logistic regression, random forest, decision tree, support vector machine, gradient boosting machine, K-nearest neighbors, linear discriminant analysis, Naive Bayes, and neural network.
[0083] III. Main Implementation Steps for Model Construction;
[0084] (a) Data collection and processing;
[0085] 1. Data collection: Collect historical disease data, including patients' clinical characteristics, laboratory test results, medical history and other information.
[0086] 2. Data Preprocessing: First, identify and remove variables with zero or near-zero variance to reduce data dimensionality and improve model training efficiency. Next, calculate the correlation matrix between variables, identify and remove highly correlated variables to avoid overfitting. Finally, assess collinearity and remove redundant variables to improve model stability and interpretability.
[0087] 3. Data partitioning: The dataset is divided into a training set (70%) and a validation set (30%) using a random partitioning method. The training set is used for model training, and the validation set is used for model hyperparameter tuning and performance evaluation.
[0088] 4. Data centralization and standardization: Centralize and standardize the data to improve the training efficiency and performance of the model.
[0089] 5. Data Feature Selection: Perform feature selection on the data to remove irrelevant or redundant features.
[0090] (II) Model Construction and Evaluation;
[0091] First, the nine models were trained using the training set data. Different hyperparameter combinations were set using either grid search or random search, and the performance of different hyperparameter combinations was evaluated using 10-fold cross-validation to determine the optimal hyperparameter combination for each of the nine machine learning models. Then, the performance of the nine machine learning models was evaluated using the validation set data to ensure the models' stability and generalization ability on different datasets. The optimal model was determined by comparing the models' accuracy and ROC. Finally, the selected optimal model was evaluated using the test set to ensure its performance in real-world applications.
[0092] (III) Practical Application;
[0093] The trained optimal model is applied to real medical data to predict the risk of thalassemia in new patients, assisting clinicians in making disease diagnosis and treatment decisions.
[0094] In one specific embodiment, blood tandem mass spectrometry data processing:
[0095] 1. Data collection: Blood tandem mass spectrometry data (56 variables in total) were collected from healthy controls and patients with intermediate thalassemia.
[0096] 2. Perform single-factor logistic regression analysis on all indicators to identify those with significant differences and then perform ROC curve analysis.
[0097] 3. The model building process is as follows: First, data preprocessing is performed:
[0098] 3a. Identify and remove zero-variance and near-zero-variance variables: There are no zero-variance and near-zero-variance variables.
[0099] 3b. Identify and remove highly correlated variables (correlation greater than 0.8): In this example, a total of 6 highly correlated variables were removed (out of a total of 56 variables, 50 variables remain).
[0100] 3c. Collinearity: There is no collinearity.
[0101] 3d. Data partitioning: The preprocessed data is randomly divided into a training set (70%) and a validation set (30%).
[0102] 3e. Data centralization and standardization: Centralizing and standardizing data processing.
[0103] 3f. Data feature selection: Lasso regression was used to select features from the remaining 11 variables.
[0104] In one specific embodiment, a one-way logistic regression analysis was performed on all indicators between the two groups. ROC curve analysis was conducted on 33 indicators showing significant differences, revealing that 9 indicators had an AUC ≥ 0.75. Among these, C4 (butyrylcarnitine) had the highest AUC at 0.941, followed by C18:1 (octadecenoylcarnitine, AUC 0.887), C5DC / C6OH (glutaryl-CoA / 3-hydroxyhexanoyl-CoA, a mixed form of a substance, the AUC is related to the detection instrument, AUC 0.877), and C16:1OH (3-hydroxyhexanoyl-CoA). The following compounds were analyzed: C1 (C16+C18) (C16+C18) (free carnitine, palmitoyl carnitine, AUC 0.872), C3 (C14+C18) (C16+C18) (the ratio of free carnitine to the sum of palmitoyl carnitine and octadecyl carnitine, AUC 0.758); and ROC curves (AUC ≥ 0.75) were plotted for indicators of significant difference in univariate logistic regression analysis. Figure 4 As shown in the figure, the sensitivity of these indicators ranges from 85.7% to 100%, and the specificity ranges from 57.6% to 84.8%. The cut-off values for each indicator in the figure do not refer to the actual measured values of each indicator, but rather to the optimal probability threshold determined by the Youden index. When the probability value of a certain indicator in logistic regression analysis is greater than or equal to the cut-off value, it suggests that the individual may have intermediate thalassemia.
[0105] In one specific embodiment, model building and evaluation:
[0106] 1. The performance of the models on the validation set is shown in Table 1 below. The results show that, except for the decision tree model, other models have better performance on the validation set (higher AUC value and higher accuracy).
[0107] Table 1 Results of ROC curves for each model
[0108] Model AUC Accuracy Sensitivity Specificity Precision P Value GBM 0.96 (0.91-1.00) 0.95 0.67 1.00 1.00 0.00 SVM 0.95 (0.88-1.00) 0.94 0.57 1.00 1.00 0.00 Tree 0.85 (0.77-0.94) 0.86 0.57 0.91 0.52 0.46 RF 0.96 (0.92-1.00) 0.92 0.48 1.00 1.00 0.01 NN 0.97 (0.93-1.00) 0.95 0.67 0.99 0.93 0.00 GLM 0.98 (0.96-1.00) 0.94 0.62 0.99 0.93 0.00 KNN 0.95 (0.89-1.00) 0.90 0.33 1.00 1.00 0.06 LDA 0.98 (0.96-1.00) 0.92 0.43 1.00 1.00 0.02 NB 0.98 (0.96-1.00) 0.95 0.67 0.99 0.93 0.00
[0109] 2. The models were evaluated, and the calibration curves showed that the GBM model had a higher goodness of fit (e.g., Figure 5 As shown in the figure), the boost curves of the GBM model are steeper and have higher initial values, indicating that the model performs better (e.g., Figure 6 As shown), the GBM model performs well at most threshold probabilities and has high clinical application value (e.g. Figure 7 (As shown). Based on the results of the ROC curve, the GBM model is currently considered the optimal model for predicting intermediate thalassemia in newborns.
[0110] 3. The relative weights of the indicators included in each model are as follows: Figure 8 As shown, C4 has the highest relative weight in the GBM model, followed by C5DC / C6OH, C0 / (C16+C18), C18:1, C3 / C2, C18OH, C5 / C2, C16:1OH, Leu / Phe, C14:1 / C16, and C8 / C10.
[0111] In one specific embodiment, the specific implementation steps for training a graph convolutional network model include data processing, domain knowledge introduction, and a hard-guided mechanism:
[0112] 1. Graph structure construction:
[0113] Node definition: Each detected mass spectrum peak is a node, and each node contains [m / z value, peak intensity, peak area, signal-to-noise ratio, peak width]; there are a total of V nodes, corresponding to V significant peaks.
[0114] Edge definition (based on domain knowledge): constructing edges between adjacent nodes based on biochemical relationships or constructing edges between adjacent nodes based on hemoglobin-specific knowledge.
[0115] Biochemical relationships include: edges formed by isotopic relationships, edges formed by charge state relationships, and edges formed by adduct relationships. Edges formed by isotopic relationships connect isotopic peaks of the same compound, with intensities conforming to isotopic distribution patterns. The criterion is an m / z difference of 1.0034 / n (where n is the charge number). Edges formed by charge state relationships connect different charge states of the same compound. The criterion is an m / z ratio equal to the charge number ratio. Edges formed by adduct relationships connect and / or connect different adduct forms of the same compound based on known mass differences of the adducts.
[0116] Edges constructed based on specific knowledge of hemoglobin include: edges constructed within hemoglobin subunits and edges constructed based on pathology-related knowledge. Edges constructed within hemoglobin subunits connect and / or link different characteristic peaks of the same hemoglobin subunit based on known hemoglobin peptide quality. Edges constructed based on pathology-related knowledge connect known hemoglobin variant peaks associated with thalassemia.
[0117] Different types of mass spectrum data are generated based on nodes and edges.
[0118] 2. Graph Convolutional Neural Network (GNN) structure, including: input layer, graph convolutional layer, graph attention layer, graph pooling layer, and classification output layer. Specifically:
[0119] The constructed mass spectrometry data is represented by node feature matrices and adjacency matrices as input. These matrices are then fed into the graph convolutional layer to aggregate the features of each node's neighbors. The aggregated node features are then fed into the graph attention layer to calculate the importance weights between nodes, allowing the model to learn the importance of different neighbors. The output of the graph attention layer is then fed into the graph pooling layer to calculate the importance of each node for the classification task. The embedding vector of the entire graph is obtained through attention pooling or global average pooling. Finally, the graph embedding vector is fed into the classification layer to calculate the classification probability.
[0120] During training, graph convolutional neural networks (CNNs) optimize their parameters using cross-entropy loss and guided loss functions. The guided loss function guides important nodes to access more relevant features and assigns higher attention scores. The loss function L... total It is expressed as follows:
[0121] L total = L classification + μL structure_guidance + νL node_guidance
[0122] Among them, L classification L represents the cross-entropy classification loss. structure_guidance This indicates that there are enough connections to guide important nodes, L node_guidance The high attention score indicates that the guiding important nodes have high attention scores, and μ and ν represent the balancing hyperparameters.
[0123] L structure_guidance = -Σ i ∈H log(Σ j ∈N(i) A ij )
[0124] L node_guidance = Σ i ∈H max(0,β- S i )²
[0125] H represents the set of nodes, β represents the expected graph attention weights corresponding to the nodes, and S i Let A represent the importance score of node i during graph pooling, where N(i) represents the aggregated node features. ij This represents the adjacency matrix, showing the connection relationship between node i and node j.
[0126] In another specific embodiment, when constructing the mass spectrum data, acylcarnitine features are added to the node feature matrix to form second mass spectrum data. The second mass spectrum data is then input into a graph convolutional neural network for intermediate thalassemia prediction. The second mass spectrum data guides the acylcarnitine features, forming graph data with multi-dimensional and multi-modal features, which provides reliability and improves accuracy for the prediction results.
[0127] Graph convolutional neural networks have excellent predictive performance for intermediate thalassemia. By directly encoding the chemical relationships between peaks in mass spectrometry data, or encoding hemoglobin relationships, or encoding acylcarnitine, nodes and edges have clear biological meanings. The constructed mass spectrometry data has rich feature dimensions, which is helpful for disease prediction.
[0128] The present invention also discloses a computer program product or system, including a computer program that, when executed by a processor, implements the above-described steps of the method for predicting intermediate thalassemia.
[0129] Figure 2 A schematic diagram of the prediction system for intermediate thalassemia provided in this embodiment of the invention specifically includes:
[0130] Acquisition Unit: Acquires tandem mass spectrometry data of the subject's blood;
[0131] Feature unit: Mass spectrometry features are obtained by feature extraction based on the blood tandem mass spectrometry data; the mass spectrometry features include: butyrylcarnitine, octadecenoylcarnitine, glutaryl-CoA, and 3-hydroxyhexanoyl-CoA;
[0132] Prediction Unit: Based on the mass spectrometry characteristics, intermediate thalassemia is predicted to obtain the prediction result.
[0133] Figure 3 An embodiment of the present invention provides a schematic diagram of a computer device, specifically including:
[0134] A memory and a processor; the memory is used to store program instructions; the processor is used to invoke the program instructions, when any of the above-mentioned predictions of intermediate thalassemia are executed.
[0135] The present invention also discloses a computer-readable storage medium storing a computer program that, when executed by a processor, provides any of the above-described predictions of intermediate thalassemia.
[0136] The verification results of this verification embodiment show that assigning inherent weights to indications can improve the performance of this method compared to the default settings. Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for example, the division of units is merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be electrical, mechanical, or other forms. The units described as separate components may or may not be physically separated; the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of this embodiment. Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk, etc.
[0137] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.
[0138] The computer device provided by the present invention has been described in detail above. For those skilled in the art, there will be changes in the specific implementation and application scope based on the ideas of the embodiments of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method of predicting intermediate Mediterranean-type anemia, characterized by, The method comprises the following steps: S1, obtaining blood tandem mass spectrometry data of a subject to be tested; S2, performing feature extraction based on the blood tandem mass spectrometry data to obtain mass spectrometry features, wherein the mass spectrometry features comprise butyryl carnitine, octadecenoyl carnitine, glutaryl coenzyme A and 3-hydroxyhexanoyl coenzyme A; S3, performing thalassemia trait prediction based on the mass spectrometry features to obtain a prediction result.
2. The method of predicting thalassemia intermedia according to claim 1, wherein The mass spectrometry features further comprise any one or more of the following: a ratio of the sum of free carnitine and palmitoyl carnitine, octadecanoyl carnitine, propionyl carnitine / acetylcarnitine, 3-hydroxyoctadecanoyl carnitine, isovalerylcarnitine / acetylcarnitine, 3-hydroxypalmitoylcarnitine, leucine / phenylalanine, myristoylcarnitine / palmitoylcarnitine, octanoylcarnitine / oleoylcarnitine, propionyl carnitine, isovalerylcarnitine, myristoylcarnitine.
3. The method of predicting thalassemia intermedia according to claim 1, wherein The prediction is performed by a prediction model, and the mass spectrometry features are input into the prediction model to perform thalassemia trait prediction and obtain a prediction result; Optionally, the prediction model comprises any one or more of the following: GBM, SVM, RF, GLM, KNN, LDA, decision tree, random forest, and naive Bayes.
4. The method of predicting thalassemia intermedia according to claim 3, wherein The prediction model construction training process comprises the following steps: Obtaining a blood tandem mass spectrometry data set; Performing variable identification on the blood tandem mass spectrometry data set, and removing variables to obtain screened variables; Performing collinearity calculation on the screened variables, and removing collinear variables to obtain second screened variables; Inputting the second screened variables into a machine learning model to be trained to perform feature learning and obtain a prediction model; Optionally, the variable identification comprises zero-variance variables, near-zero-variance variables, and highly-correlated variables.
5. The method of predicting thalassemia intermedia according to claim 1, wherein S2 is replaced by: inputting the blood tandem mass spectrometry data into a second prediction model to perform thalassemia trait prediction and obtain a prediction result, wherein the second prediction model is obtained by neural network training; and S3 is removed; Optionally, the neural network comprises any one or more of the following: one-dimensional convolutional neural network, Transformer, and graph neural network.
6. The method of predicting thalassemia intermedia according to claim 5, wherein The neural network is a graph neural network, which is obtained by constructing graph data from blood tandem mass spectrometry data and training the graph data and domain knowledge; Optionally, the training process of the graph neural network comprises the following steps: Obtaining a blood tandem mass spectrometry data set; Based on the blood tandem mass spectrometry data set, each peak detected in the mass spectrometry data is taken as a node, and edges are constructed between the nodes through biochemical relationships between the nodes to obtain mass spectrometry graph data; Inputting the mass spectrometry graph data into a graph neural network to be trained to obtain the graph neural network; Optionally, the biochemical relationship comprises any one or more of the following: isotopic relationship, charge relationship, and adduct relationship, and the edges are constructed through the isotopic relationship and / or the charge relationship and / or the adduct relationship between the nodes; Optionally, the biochemical relationship is replaced by a hemoglobin relationship, and the edges are constructed through the hemoglobin relationship between the nodes; Optionally, the hemoglobin relationship comprises different feature peaks of the same hemoglobin subunit, hemoglobin peptide segment mass, and / or hemoglobin variant peak; The hemoglobin variant peak is a hemoglobin variant peak known to be associated with thalassemia. Optionally, the construction edge further comprises a hemoglobin relationship, and the edge is constructed between the nodes through the biochemical relationship and the hemoglobin relationship.
7. The method of predicting thalassemia intermedia according to claim 6, wherein The node in the mass spectrum data further comprises acylcarnitine, the mass spectrum data comprises a node feature matrix and an adjacency matrix, a second node feature matrix is obtained by adding an acylcarnitine feature to the node feature matrix constructed by the node, the second node feature matrix and the adjacency matrix constructed by the edge form second mass spectrum data, and the second mass spectrum data is input into the graph neural network to be trained to obtain the graph neural network. Optionally, the acylcarnitine comprises any one or more of the following: butyrylcarnitine, octadecenoylcarnitine, glutaryl-CoA\3-hydroxyhexanoyl-CoA, the ratio of the sum of free carnitine and palmitoylcarnitine, propionylcarnitine / acetylcarnitine, 3-hydroxyoctadecanoylcarnitine, isovalerylcarnitine / acetylcarnitine, 3-hydroxypalmitoylcarnitine, leucine / phenylalanine, myristoylcarnitine / palmitoylcarnitine, octanoylcarnitine / phytanic acid, propionylcarnitine, isovalerylcarnitine, myristoylcarnitine.
8. A computer program product comprising a computer program or instructions embodied therein, characterized in that, The computer program or instruction is executed by the processor to realize the prediction method of the thalassemia intermedia according to any one of claims 1-7.
9. A computer device comprising a memory, a processor, and a computer program or instructions stored on the memory, wherein, The computer program or instruction is executed by the processor to realize the prediction method of the thalassemia intermedia according to any one of claims 1-7.
10. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instruction is executed by the processor to realize the prediction method of the thalassemia intermedia according to any one of claims 1-7. The computer program or instruction is executed by the processor to realize the prediction method of the thalassemia intermedia according to any one of claims 1-7.