Device for determining the relationship between anti-double-stranded DNA IgG glycosylation and organ involvement in systemic lupus erythematosus

Abstract

The present invention relates to a device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement, comprising: constructing a coarse-grained classification model, training the coarse-grained classification model using pairs of anti-double-stranded DNA IgG glycoforms, evaluating multiple trained coarse-grained classification models, and identifying several top-ranked trained coarse-grained classification models from the evaluation results; constructing a fine-grained classification model, training the fine-grained classification model using inputs corresponding to the top-ranked trained coarse-grained classification models, evaluating the multiple trained fine-grained classification models, identifying the trained fine-grained classification model with the best evaluation results, and using the input of the trained fine-grained classification model with the best evaluation results as a primary indicator for predicting systemic lupus erythematosus organ involvement. The present invention can determine the relationship between anti-double-stranded DNA IgG glycosylation and SLE clinical manifestations.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence-assisted medical technology, and in particular to a device for determining the relationship between anti-double-stranded DNA IgG glycosylation level and systemic lupus erythematosus. Background Art

[0002] Systemic lupus erythematosus (SLE) is a multifaceted autoimmune disease with diverse clinical manifestations and multi-organ system involvement. The heterogeneity of this disease presents significant challenges for assessment and treatment, requiring a detailed understanding of the underlying pathophysiological mechanisms. Recent advances in immunology have revealed a key role for immunoglobulin G (IgG) glycosylation in the pathogenesis of SLE, providing potential avenues for the development of more precise diagnostic and prognostic tools. Glycosylation is a post-translational modification that can significantly influence the biological properties of antibodies, including their effector functions, stability, and interactions with immune cells. In SLE, altered IgG glycosylation patterns have been observed and may be associated with disease pathogenesis. For example, alterations in IgG Fc glycosylation, particularly decreased galactosylation and increased focalization, have been found to be associated with disease activity and progression in SLE. Within the broader context of rheumatic diseases, glycosylation has been studied in rheumatoid arthritis (RA), where it is known to influence antibody and immune cell function. For example, the glycosylation signature of the IgG Fc domain in RA is a shift toward galactosylation, which is associated with increased inflammation and disease severity. This finding suggests that glycosylation changes may play a common role in the pathogenesis of different rheumatic diseases and may affect disease progression and organ-specific manifestations.

[0003] The relationship between IgG glycosylation and organ involvement in SLE is an area of ​​active research. It is hypothesized that altered glycosylation may contribute to the deposition of immune complexes in target organs, leading to inflammation and tissue damage. Fucosylation, galactosylation, dichotomization, and sialylation of IgG subclasses play a crucial role in this process. For example, IgG glycosylation patterns have been associated with the severity of nephritis in patients with SLE. Furthermore, IgG glycosylation profiles have emerged as promising biomarkers for SLE, with distinct glycosylation patterns associated with distinct disease outcomes. Anti-double-stranded DNA IgG is a specific antibody for SLE, and its glycosylation may influence its pathogenic potential in SLE. Aberrant glycosylation can modulate the ability of these antibodies to activate the complement system, bind to Fcγ receptors on immune cells, and form immune complexes, which are associated with the development of lupus nephritis. Furthermore, the glycosylation status of anti-double-stranded DNA IgG may serve as a biomarker for SLE disease activity and organ involvement.

[0004] However, the relationship between anti-dsDNA antibody glycosylation and SLE organ involvement remains unclear, and currently there is no effective method to evaluate and predict SLE organ involvement based on anti-dsDNA antibody glycosylation levels. The complexity of these glycosylation patterns and their interaction with clinical symptoms require sophisticated analytical methods to fully elucidate their significance. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement, which can find the relationship between anti-double-stranded DNA IgG glycosylation and SLE clinical manifestations.

[0006] The technical solution adopted by the present invention to solve the technical problem is to provide a device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement, comprising:

[0007] The first construction module is used to construct a coarse-grained classification model, wherein the input of the coarse-grained classification model is a combination of different anti-double-stranded DNA IgG glycoforms, and the output is the degree of organ system involvement;

[0008] A first training module is used to train the coarse-grained classification model using anti-double-stranded DNA IgG glycoforms in pairs to obtain multiple trained coarse-grained classification models, evaluate the multiple trained coarse-grained classification models, and find several trained coarse-grained classification models with high rankings from the evaluation results;

[0009] A second construction module is used to construct a fine-grained classification model, the input of the fine-grained classification model is the input corresponding to the top-ranked trained coarse-grained classification models, and the output is the organ system involvement score;

[0010] The second training module is used to train the fine-grained classification model using the inputs corresponding to the top-ranked trained coarse-grained classification models to obtain several trained fine-grained classification models, and to evaluate the several trained fine-grained classification models to find the trained fine-grained classification model with the best evaluation result;

[0011] The indicator determination module is used to use the input of the trained fine-grained classification model with the best evaluation result as the main indicator for predicting organ involvement in systemic lupus erythematosus.

[0012] The coarse-grained classification model is a random forest model.

[0013] The anti-double-stranded DNA IgG glycoforms include IgG3 / 4Bis, IgG1Gal, IgG1Fuc, IgG2Sia, IgG2Gal, IgG3 / 4Sia, IgG3 / 4Gal, IgG2Fuc, IgG1Sia, IgG3 / 4Fuc, IgG2Bis and IgG1Bis.

[0014] When the first training module uses pairwise combinations of anti-double-stranded DNA IgG glycoforms to train the coarse-grained classification model, each combination trains the coarse-grained classification model K times, and the average AUC of the coarse-grained classification model trained K times is used as the evaluation indicator of the trained coarse-grained classification model.

[0015] The fine-grained classification model is an artificial neural network model.

[0016] When the second training module evaluates the several trained fine-grained classification models, mean square error is used as the evaluation indicator of the several trained fine-grained classification models.

[0017] The device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement also includes:

[0018] a third training module, configured to combine the anti-double-stranded DNA IgG glycoforms in the input corresponding to the top-ranked trained coarse-grained classification models into three or four combinations, and train the coarse-grained classification models using the combined anti-double-stranded DNA IgG glycoforms to obtain N trained coarse-grained classification models, and evaluate the N trained coarse-grained classification models;

[0019] A judgment module is used to judge whether the coarse-grained classification model with the best evaluation result among N trained coarse-grained classification models is better than the coarse-grained classification model with the best evaluation result among several trained coarse-grained classification models;

[0020] a first determination module, configured to, when a coarse-grained classification model with the best evaluation result among the N trained coarse-grained classification models is better than a coarse-grained classification model with the best evaluation result among several trained coarse-grained classification models, use the coarse-grained classification model with the best evaluation result among the N trained coarse-grained classification models as the final classification model;

[0021] The second determination module is used to use the coarse-grained classification model with the best evaluation results among N trained coarse-grained classification models as the final classification model when the coarse-grained classification model with the best evaluation results among several trained coarse-grained classification models is not better than the coarse-grained classification model with the best evaluation results among several trained coarse-grained classification models.

[0022] Beneficial effects

[0023] Due to the adoption of the above-mentioned technical solution, the present invention has the following advantages and positive effects compared with the existing technology: the present invention uses a relatively non-invasive examination to identify the relationship between anti-double-stranded DNA IgG glycosylation and the clinical manifestations of SLE, thereby accurately assessing the systemic involvement of SLE patients, helping doctors to assess the patient's condition early, intervene in time, formulate appropriate treatment plans, and improve the patient's prognosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of a device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement according to an embodiment of the present invention;

[0025] Figure 2 Schematic diagram of the AUC value of the coarse-grained model corresponding to each anti-double-stranded DNA IgG glycoform in an embodiment of the present invention;

[0026] Figure 3 Schematic diagram of the results of the combination of anti-double-stranded DNA IgG glycoforms in the random forest classification model according to an embodiment of the present invention;

[0027] Figure 4 4 is a diagram showing the prediction result of the degree of organ involvement by the artificial neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] Below in conjunction with specific embodiment, further set forth the present invention.Should be understood that these embodiments are only used to illustrate the present invention and are not used in limiting the scope of the present invention.In addition, should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall equally within the scope limited by the appended claims of the application.

[0029] The embodiments of the present invention relate to a device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement, such as Figure 1 As shown, it includes: a first construction module, a first training module, a second construction module, a second training module and an indicator determination module.

[0030] The first construction module is used to construct a coarse-grained classification model, the input of which is a combination of different anti-double-stranded DNA IgG glycoforms, and the output is the degree of organ system involvement. The first training module is used to train the coarse-grained classification model using pairwise combinations of anti-double-stranded DNA IgG glycoforms to obtain multiple trained coarse-grained classification models, evaluate the multiple trained coarse-grained classification models, and identify the top-ranked trained coarse-grained classification models from the evaluation results.

[0031] In this embodiment, the coarse-grained classification model uses a random forest model. The random forest model is a machine learning algorithm widely used for classification tasks. It trains multiple decision trees and combines their integrated outputs to determine which class the sample should be assigned to. In this embodiment, the random forest algorithm is executed through the scikit-learn module on python 3.10. To make the results more convincing, the samples are divided into two groups in this embodiment, 70% for the training set and the remaining 30% for the test set. In order to improve the performance of the random forest model, the bootstrap technology is also used when sampling the data set. Different types of glycosylation of anti-double-stranded DNA IgG subclasses, including fucosylation (Fuc), galactosylation (Gal), bispecification (Bis) and sialylation (Sia), are used as classification criteria, and the degree of organ involvement (HIGH or LOW) is set as the output result of the random forest model. Among them, the HIGH group represents patients with 3 or more organ systems involved, and the LOW group represents patients with 2 or fewer organ systems involved.

[0032] In order to verify the effectiveness of the model and compare the performance of different anti-dsDNA IgG glycoforms, the present embodiment applies the model to samples in the test set. Then, the area under the receiver operating characteristic (ROC) curve (AUC) and the classification accuracy of the test set are calculated as evaluation means. It is worth noting that, instead of using a single anti-dsDNA IgG glycoform of a single IgG subclass as input each time, the present embodiment uses a pair of anti-dsDNA IgG glycoforms each time to improve the coarse-grained classification model. A pair of different anti-dsDNA IgG glycoforms generally obtains a higher AUC value than using one of the anti-dsDNA IgG glycoforms alone, which indicates that the combination of two anti-dsDNA IgG glycoforms is better than using one anti-dsDNA IgG glycoform alone, because a pair of anti-dsDNA IgG glycoforms carries more information than using one anti-dsDNA IgG glycoform alone.

[0033] In this embodiment, a coarse-grained classification model is trained 100 times for each anti-double-stranded DNA IgG glycoform, and the average AUC of the 100 coarse-grained classification models is used as the AUC of the coarse-grained classification model corresponding to the anti-double-stranded DNA IgG glycoform. The anti-double-stranded DNA IgG glycoforms are sorted in descending order based on the average AUC value obtained by combining them with all 12 anti-double-stranded DNA IgG glycoforms. Figure 2 As shown, the classification ability from front to back is: IgG3 / 4Bis, IgG1Gal, IgG1Fuc, IgG2Sia, IgG2Gal, IgG3 / 4Sia, IgG3 / 4Gal, IgG2Fuc, IgG1Sia, IgG3 / 4Fuc, IgG2Bis, IgG1Bis. Figure 2 A clear trend can be seen in the results: higher-ranked anti-dsDNA IgG glycoform combinations generally exhibited better AUC values ​​than lower-ranked anti-dsDNA IgG glycoform combinations. Furthermore, some anti-dsDNA IgG glycoforms exhibited high AUC values ​​when combined with most other anti-dsDNA IgG glycoforms, indicating that they do have a significant impact on the severity of organ system involvement in patients. The average AUC for the anti-dsDNA IgG glycoform pair IgG1Gal & IgG3 / 4Bis reached 0.6300, exceeding that of other anti-dsDNA IgG glycoform pairs, indicating that the IgG1Gal & IgG3 / 4Bis glycoform pair best correlated with SLE organ involvement. Meanwhile, several other lower-ranked anti-dsDNA IgG glycoform pairs (e.g., IgG1Bis & IgG3 / 4Fuc, IgG1Bis & IgG2Bis) exhibited AUC values ​​below 0.5, indicating that they were unable to categorize the severity of organ involvement.

[0034] The apparatus for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement in this embodiment may further include: a third training module, the third training module being used to train the anti-double-stranded DNA in the input corresponding to the top-ranked coarse-grained classification models; The IgG glycoforms are combined into three or four combinations, and the combined anti-double-stranded DNA IgG glycoforms are used to train the coarse-grained classification model to obtain N trained coarse-grained classification models, and the N trained coarse-grained classification models are evaluated; a judgment module is used to judge whether the coarse-grained classification model with the best evaluation results among the N trained coarse-grained classification models is better than the coarse-grained classification model with the best evaluation results among several trained coarse-grained classification models; a first determination module is used to use the coarse-grained classification model with the best evaluation results among the N trained coarse-grained classification models as the final classification model when the coarse-grained classification model with the best evaluation results among the N trained coarse-grained classification models is better than the coarse-grained classification model with the best evaluation results among several trained coarse-grained classification models; a second determination module is used to use the coarse-grained classification model with the best evaluation results among several trained coarse-grained classification models as the final classification model when the coarse-grained classification model with the best evaluation results among the N trained coarse-grained classification models is not better than the coarse-grained classification model with the best evaluation results among several trained coarse-grained classification models.

[0035] After obtaining the classification ability ranking of each anti-dsDNA IgG glycoform, several top-ranked anti-dsDNA IgG glycoforms are selected and recombined in groups of three and four. The RF algorithm is then re-implemented to generate an RF model with optimal classification performance. This process yields N newly trained coarse-grained classification models, which are evaluated by their classification accuracy and AUC, providing evidence for the glycoform combination that best characterizes the degree of organ involvement in SLE. This embodiment tested various combinations, and the classification accuracy of combinations of three and four anti-dsDNA IgG glycoforms was comparable to that of the sugar pairs. Furthermore, combinations of more than three glycoforms lacked significance. Therefore, this embodiment uses the coarse-grained classification model with the best evaluation results among the coarse-grained classification models trained in the first training module as the final classification model.

[0036] This embodiment studies and compares the classification ability of sugar pairs by implementing the RF algorithm. Among a few sugar pairs, when the RF model is applied to the test set, the anti-double-stranded DNA IgG1Gal&IgG3 / 4Bis and IgG1Fuc&IgG3 / 4Bis combinations (see Figure 3 a and c) in the above table have the highest classification accuracy for the degree of organ involvement, followed by the IgG1Fuc&IgG2Sia and IgG2Sia&IgG3 / 4Bis combinations (see Figure 3 d and f in ). However, due to the limited size of the test set, the accuracy was unable to distinguish certain anti-dsDNA IgG glycoform combinations. To further validate the results, the AUC was calculated to make the classification performance more reliable. Figure 3 In g and l, the corresponding ROC curves with the highest AUC values ​​among all trained RF models are plotted, and the 95% CI is given to demonstrate the effectiveness of each anti-dsDNA IgG glycoform pair. Figure 3 g) also has the highest AUC value (0.8187) and 95% confidence interval (0.6111, 0.6388). Anti-double-stranded DNA IgG1Fuc&IgG3 / 4Bis ( Figure 2 i) and IgG1Gal&IgG3 / 4Bis( Figure 2 a) in has the same highest accuracy ( Figure 3 c), but the AUC value was lower (0.7820), 95% CI (0.5204, 0.5673). Other combinations such as IgG1Gal&IgG2Gal and IgG2Sia&IgG3 / 4Bis ( Figure 3The k and 1) in the classification of organ system involvement were also slightly weaker. Among the two evaluation indicators, anti-double-stranded DNA IgG1Gal & IgG3 / 4Bis performed best in the classification task. Moreover, these two anti-double-stranded DNA IgG glycoforms were Figure 2 This indicates that they are indeed strongly correlated with the degree of involvement, and their combination can be used as an important indicator of SLE organ involvement.

[0037] The second construction module in this embodiment is used to construct a fine-grained classification model, the input of the fine-grained classification model is the input corresponding to the top-ranked trained coarse-grained classification models, and the output is the organ system involvement score; the second training module is used to train the fine-grained classification model using the input corresponding to the top-ranked trained coarse-grained classification models to obtain several trained fine-grained classification models, and evaluate the several trained fine-grained classification models to find the trained fine-grained classification model with the best evaluation result; the indicator determination module is used to use the input of the trained fine-grained classification model with the best evaluation result as the main indicator for predicting organ involvement of systemic lupus erythematosus.

[0038] The fine-grained classification model of this embodiment can be an artificial neural network (ANN) model, which can define the degree of involvement as a continuous value between 0 and 1. A patient with x number of involved organs is assigned an involvement rate of x / 7. Since the number of organ systems involved in the actual sample can only be an integer, the actual involvement rate has only a few possible values. However, the predicted involvement rate can be any value between 0 and 1, reflecting the severity of the impact on multiple organ systems. It was finally found that the ANN model of anti-double-stranded DNA IgG1Gal&IgG3 / 4Bis had the lowest MSE value among all sugar combinations (see Figure 4 Compared with other sugar combinations, the red and blue lines have more similar trends. The results show that when the model was trained with anti-dsDNA IgG1Gal & IgG3 / 4Bis, the difference between predicted and actual values ​​was minimal, and it was able to effectively predict the severity of organ involvement, which is consistent with previous classification results. Therefore, anti-dsDNA IgG1Gal & IgG3 / 4Bis was selected as the primary indicator for predicting organ involvement in systemic SLE.

[0039] It is not difficult to find that the present invention uses a relatively non-invasive examination to find the relationship between anti-double-stranded DNA IgG glycosylation and the clinical manifestations of SLE, thereby being able to correctly assess the systemic involvement of SLE patients, helping doctors to assess the patient's condition early, intervene in time, formulate appropriate treatment plans, and improve patient prognosis.

Claims

1. A device for determining the relationship between anti-double-stranded DNA IgG glycosylation and organ involvement in systemic lupus erythematosus, characterized in that: include: The first construction module is used to construct a coarse-grained classification model, wherein the input of the coarse-grained classification model is a combination of different anti-double-stranded DNA IgG glycoforms, and the output is the degree of organ system involvement; the coarse-grained classification model is a random forest model; The first training module is used to train the coarse-grained classification model using anti-double-stranded DNA IgG glycoforms in pairs to obtain multiple trained coarse-grained classification models, and evaluate the multiple trained coarse-grained classification models, and find several trained coarse-grained classification models with high rankings from the evaluation results; A second construction module is configured to construct a fine-grained classification model, wherein the input of the fine-grained classification model is the input corresponding to the top-ranked trained coarse-grained classification models, and the output is the organ system involvement score; the fine-grained classification model is an artificial neural network model; The second training module is used to train the fine-grained classification model using the inputs corresponding to the top-ranked trained coarse-grained classification models to obtain several trained fine-grained classification models, and to evaluate the several trained fine-grained classification models to find the trained fine-grained classification model with the best evaluation result; The indicator determination module is used to use the input of the trained fine-grained classification model with the best evaluation result as the main indicator for predicting organ involvement in systemic lupus erythematosus.

2. The device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement according to claim 1, characterized in that: The anti-double-stranded DNA antibody IgG glycoforms include IgG3 / 4Bis, IgG1Gal, IgG1Fuc, IgG2Sia, IgG2Gal, IgG3 / 4Sia, IgG3 / 4Gal, IgG2Fuc, IgG1Sia, IgG3 / 4Fuc, IgG2Bis and IgG1Bis.

3. The device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement according to claim 1, characterized in that: When the first training module uses pairwise combinations of anti-double-stranded DNA IgG glycoforms to train the coarse-grained classification model, each combination trains the coarse-grained classification model K times, and the average AUC of the coarse-grained classification model trained K times is used as the evaluation indicator of the trained coarse-grained classification model.

4. The device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement according to claim 1, characterized in that: When the second training module evaluates the several trained fine-grained classification models, mean square error is used as the evaluation indicator of the several trained fine-grained classification models.

5. The device for determining the relationship between anti-double-stranded DNA IgG glycosylation and systemic lupus erythematosus organ involvement according to claim 1, characterized in that: Also includes: a third training module, configured to combine the anti-double-stranded DNA IgG glycoforms in the input corresponding to the top-ranked trained coarse-grained classification models into three or four combinations, and train the coarse-grained classification models using the combined anti-double-stranded DNA IgG glycoforms to obtain N trained coarse-grained classification models, and evaluate the N trained coarse-grained classification models; A judgment module is used to judge whether the coarse-grained classification model with the best evaluation result among N trained coarse-grained classification models is better than the coarse-grained classification model with the best evaluation result among several trained coarse-grained classification models; a first determination module, configured to, when a coarse-grained classification model with the best evaluation result among the N trained coarse-grained classification models is better than a coarse-grained classification model with the best evaluation result among several trained coarse-grained classification models, use the coarse-grained classification model with the best evaluation result among the N trained coarse-grained classification models as the final classification model; The second determination module is used to use the coarse-grained classification model with the best evaluation results among N trained coarse-grained classification models as the final classification model when the coarse-grained classification model with the best evaluation results among several trained coarse-grained classification models is not better than the coarse-grained classification model with the best evaluation results among several trained coarse-grained classification models.

Images