Ankylosing spondylitis prediction model construction method based on neural network

Abstract

The invention discloses an ankylosing spondylitis prediction model construction method based on a neural network, and belongs to the technical field of biomedicine, and the method specifically comprises the following steps: sample collection: collecting a blood sample; preparing a sample; performing mass spectrum identification: performing mass spectrum analysis by adopting a Linear mode of a mass spectrometer; performing data analysis; performing MS file processing; performing mass spectrum detection on peak profile data, and setting a threshold value as Plt; 0.01 and log2 (FC) gt; 1, screening a differential mass spectrum peak; carrying out LASSO dimension reduction on the differential mass spectrum peaks to screen mass spectrum characteristic peaks; analyzing the identification efficiency of the mass spectrum characteristic peak on the ankylosing spondylitis by an ROC method; carrying out artificial neural network algorithm modeling; and analyzing and predicting the data of the test set by using the artificial neural network model. More detailed and comprehensive biomarker analysis can be provided, and the overall metabolic state and the pathological process of a patient and good medication curative effect evaluation can be reflected.

Description

Technical Field

[0001] The invention belongs to the field of biomedical technology, and in particular relates to a method for constructing a prediction model of ankylosing spondylitis based on a neural network. Background Art

[0002] Ankylosing Spondylitis (AS) is an autoimmune disease characterized by chronic inflammatory low back pain, morning stiffness, and discomfort in the hips and groin. Its pathological features include involvement of the sacroiliac joints, bone erosion, and inflammation of the soft tissues around the joints. As the disease progresses, the patient's spine may undergo fibrosis, causing gradual loss of function, eventually presenting a "bamboo-like" change, and may lead to severe disability. In addition, AS may also affect peripheral joints and be accompanied by other manifestations, such as inflammatory bowel disease, uveitis, and pulmonary fibrosis. At present, the exact cause and pathogenesis of AS are still unclear, and there is no cure. Existing treatments mainly focus on relieving symptoms and restoring function. Drug treatments include nonsteroidal anti-inflammatory drugs, immunosuppressants, and glucocorticoids. However, long-term use of these drugs may lead to many side effects, such as gastrointestinal problems, liver and kidney damage, and increased risk of cardiovascular and cerebrovascular diseases. Biological agents, such as tumor necrosis factor (TNF) α inhibitors and interleukin (IL)-17 inhibitors, can effectively control the disease, but they also bring risks of infection, bone marrow suppression and malignant tumors. In recent years, the study of early predictive markers for AS has gradually attracted attention, especially the application of non-targeted metabolites combined with machine learning technology has provided a new perspective. Non-targeted metabolomics can analyze metabolites in patients' blood or urine, and reveal biomarkers related to the development of AS through data mining and pattern recognition. This new method not only helps to identify high-risk patients early, but also can be predicted through machine learning models, thereby achieving faster and more efficient screening. Although the current research on metabolites as early predictive markers is still in development, its potential cannot be ignored.

[0003] Therefore, establishing an early prediction and treatment strategy based on plasma molecular fingerprint and machine learning is of great significance for the early detection and treatment prognosis and efficacy evaluation of AS patients. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for constructing a prediction model for ankylosing spondylitis based on a neural network in response to the shortcomings of the background technology, which can provide a more detailed and comprehensive analysis of biomarkers, reflect the patient's overall metabolic state and pathological process and the efficacy evaluation after medication.

[0005] The present invention adopts the following technical solutions to solve the above technical problems:

[0006] A method for constructing a prediction model for ankylosing spondylitis based on a neural network, specifically comprising the following steps;

[0007] Step 1, specimen collection and processing preparation;

[0008] Step 2, mass spectrometry identification: mass spectrometry analysis is performed using the linear mode of the mass spectrometer;

[0009] Step 3, data analysis: perform LASSO dimensionality reduction on the differential mass spectrum peaks to screen the mass spectrum characteristic peaks;

[0010] Step 4: Artificial neural network modeling based on mass spectrum characteristic peaks: The calculation process is shown in the following formula:

[0011] A. Error term of the output layer

[0012]

[0013] in, is the output of the jth neuron in the output layer, y j is the target value, σ' is the derivative of the activation function; is the weighted input of the activation function of the jth neuron in the Lth layer;

[0014] B. Error term of hidden layer

[0015]

[0016] in, is the derivative of the activation function, is the weight from the jth neuron in layer I to the kth neuron in layer I+1, is the error term of the I+1th layer, is the output of the kth neuron in layer I;

[0017] C. Gradient of loss function with respect to bias

[0018]

[0019] in, is the error term of the jth neuron in layer I;

[0020] D. Gradient of loss function with respect to weights

[0021]

[0022] in, is the input of the kth neuron in layer I.

[0023] As a further preferred embodiment of the method for constructing a prediction model of ankylosing spondylitis based on a neural network of the present invention, in step 1, a blood sample is collected; the blood sample comprises 48 blood samples, including 16 samples in a healthy group and 32 samples in an AS group; after the sample is collected, it is centrifuged at 2400g×10min to separate the plasma and stored at -80°C.

[0024] As a further preferred embodiment of the method for constructing a prediction model of ankylosing spondylitis based on a neural network of the present invention, in step 1, blood sample preparation: plasma sample (25 μl) is vigorously vortexed for 10 min at room temperature with 100 μl of a methanol / chloroform mixture (2:1, v / v), centrifuged at 6000 g for 10 min at 4°C, and the supernatant is collected; the supernatant sample is diluted 10 times with sterile deionized water, vortexed and mixed, 1 μl of the sample is aspirated and spotted on a metal target plate, and 1 μl of MS-CHCA matrix containing 31.0% (m / v) of α-cyano-4-hydroxy-cinnamic acid, 25.57% (m / v) of ethanol and 25.4% (m / v) of acetonitrile is added at the same time; and the sample is dried at room temperature for 10 min and then tested on a machine.

[0025] As a further preferred embodiment of the method for constructing a prediction model of ankylosing spondylitis based on a neural network of the present invention, in step 2, the linear mode of the Axima Assurance mass spectrometer of Shimadzu Corporation of Japan is used for mass spectrometry analysis; the parameters are as follows: detection method: positive ion; precursor ion scanning range: 1-20000 m / z; laser intensity: 79; excitation frequency: 100 times / sample; referring to "JJF1528-2015 Calibration Specification for Time-of-Flight Mass Spectrometer" and "YYT 1740.2-2021 Medical Mass Spectrometer Part 2: Matrix-Assisted Laser Desorption Ionization Time-of-Flight Mass Spectrometer", trypsin inhibitor (6512 m / z), myohemoglobin (M 2+ The mass-to-charge accuracy, stability, resolution and repeatability of the instrument were calibrated using a mixed protein standard of 1,4-dihydropyrimidine (DTP) (8476 m / z) and cytochrome C (12355 m / z).

[0026] As a further preferred embodiment of the method for constructing a prediction model for ankylosing spondylitis based on a neural network of the present invention, in step 3,

[0027] Basic characteristics of the subjects: 16 healthy individuals and 32 AS patients were included; HLA-B27 tests in the healthy individuals group were all negative; HLA-B27 tests in the AS group were all positive;

[0028] Qualitative analysis: Plasma samples were tested by MALDI-TOF MS, and 188,363 profile peak data were obtained for each sample. PCA and UMAP dimensionality reduction analysis based on profile peak data showed differences between the two groups.

[0029] There were significant differences in mass spectrometry peaks between the two groups: the difference mass spectrometry peak DMP was defined as the mass spectrometry peak with significant difference between the two groups; a total of 3234 DMPs were found between the two groups, of which 105 were upregulated and 3129 were downregulated; the calculation formula is as follows:

[0030] DMP=p<0.01∩|log2FC|>1

[0031] Where p is the significant statistic of the comparison of the mass spectrometry peak values ​​between the disease group and the control group, and FC is the ratio of the mean mass spectrometry peak values ​​between the disease group and the control group;

[0032] In order to further screen the mass spectrometry characteristic peaks, the mass spectrometry characteristic peaks were screened by LASSO regression algorithm. The calculation process is shown in the following formula:

[0033]

[0034] Where J(θ) is the objective function of Lasso regression; MSE(θ) is the mean squared error; α is the regularization parameter, by controlling a, the overall regression coefficient can be compressed; θ1, θ2, ...θ n are the parameters of the model;

[0035] LASSO regression screening obtained 16 DMPs with mass-to-charge ratios of M / Z 92.2216, 178.5946, 207.7411, 208.4742, 342.9619, 386.2689, 1397.8525, 3152.613, 3154.0399, 8356.3611, 8545.0171, 8565.3403, 8609.5261, 8714.1381, 9534.0368 and 10165.3102 as mass spectral characteristic peaks.

[0036] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0037] The present invention discloses a method for constructing a prediction model of ankylosing spondylitis based on a neural network. The characteristic peaks of plasma molecular mass spectrometry of ankylosing spondylitis patients are studied based on MALDI-TOFMS technology, and a prediction model of ankylosing spondylitis is established through an artificial neural network algorithm. The plasma molecular fingerprint prediction model based on machine learning can provide a more detailed and comprehensive biomarker analysis, reflecting the patient's overall metabolic state and pathological process and the efficacy evaluation after medication. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 are the mass spectrum peaks of NC and AS of the present invention; wherein, upper: AS; lower: NC;

[0040] Figure 2 The volcano plot of differentially expressed molecules between NC and AS of the present invention; wherein the horizontal axis is the multiple of the expression level log2, and the vertical axis is the P value log10. Each point represents a mass spectrometry peak; red points and green points represent up-regulated mass spectrometry peaks and down-regulated mass spectrometry peaks, respectively; gray points represent mass spectrometry peaks with no difference;

[0041] Figure 3 This is a heat map of the differential mass spectrometry peaks between NC and AS of the present invention, red represents upregulated mass spectrometry peaks, and blue represents downregulated mass spectrometry peaks;

[0042] Figure 4 The LASSO regression analysis of the present invention is used to reduce the dimension of the difference mass spectrum peaks;

[0043] Figure 5 The ROC curve of mass spectrometry characteristic peak identification AS of the present invention;

[0044] Figure 6 A topological structure diagram of the training set data modeled by the artificial neural network algorithm of the present invention;

[0045] Figure 7 This is the predicted ROC curve of the algorithm model of the present invention for the validation set data. DETAILED DESCRIPTION

[0046] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings:

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. The present invention is described in detail below based on the drawings and preferred embodiments, and the purpose and effect of the present invention will become clearer. It should be understood that the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0048] The present invention can be better understood according to the following examples. However, the examples are only used to illustrate the present invention, and should not and will not limit the present invention described in detail in the claims.

[0049] Embodiment 1:

[0050] 1 Materials and instruments:

[0051] 1.1 Materials:

[0052] Blood samples were collected from outpatients and inpatients who visited the Department of Rheumatology and Immunology of Jiangsu Provincial Hospital of Traditional Chinese Medicine from May to December 2020. AS patients predicted according to the Revised New York Criteria for Ankylosing Spondylitis were included in the AS group. Healthy individuals who underwent physical examinations during the same period were included in the normal control group. Finally, 8 men and 8 women with an average age of 41.82±15.52 years were included in the healthy group of this study; 17 men and 15 women with an average age of 45.86±16.90 years were included in the AS group. This study was approved by the Ethics Committee of the Affiliated Hospital of Nanjing University of Chinese Medicine (2023NL-135-02).

[0053] 1.2 Main instruments:

[0054] Axima Assurance mass spectrometer (Shimadzu, Japan); low-temperature high-speed centrifuge (Allegra X-15R, BECKMANCOULTER, USA); electronic balance (AX324Z, OHAUS, USA); compact constant temperature mixer (HCM-100pro, Beijing Dalong Technology Co., Ltd.); vortex oscillator (GENIE vortex-2, Scientific Industries, USA).

[0055] 2 Methods:

[0056] 2.1 Specimen collection:

[0057] A total of 48 blood specimens were collected in this study, including 16 in the healthy group and 32 in the AS group. After sample collection, plasma was separated by centrifugation at 2400 g for 10 min and stored at -80 °C. All methods were performed in accordance with relevant guidelines and regulations and approved by the Ethics Committee of the Affiliated Hospital of Nanjing University of Chinese Medicine (2023NL-135-02).

[0058] 2.2 Mass spectrometry (MS):

[0059] 2.2.1 Sample preparation:

[0060] Plasma samples (25 μl) were vigorously vortexed for 10 min at room temperature with 100 μl of methanol / chloroform mixture (2:1, v / v), centrifuged at 6000 g for 10 min at 4 °C, and the supernatant was collected. The supernatant sample was diluted 10 times with sterile deionized water, vortexed and mixed, and 1 μl of the sample was pipetted onto a metal target plate for spotting, and 1 μl of MS-CHCA matrix containing α-cyano-4-hydroxy-cinnamic acid 31.0% (m / v), ethanol 25.57% (m / v) and acetonitrile 25.4% (m / v) was added at the same time. After drying at room temperature for 10 min, it was tested on the machine.

[0061] 2.2.2 Mass spectrometry identification:

[0062] The linear mode of the Axima Assurance mass spectrometer from Shimadzu Corporation of Japan was used for mass spectrometry analysis. The parameters were as follows: detection method: positive ion; precursor ion scanning range: 1-20000 m / z; laser intensity: 79; excitation frequency: 100 times / sample. Referring to the "JJF1528-2015 Calibration Specification for Time-of-Flight Mass Spectrometer" and "YYT 1740.2-2021 Medical Mass Spectrometer Part 2: Matrix-Assisted Laser Desorption Ionization Time-of-Flight Mass Spectrometer", trypsin inhibitor (6512 m / z), myohemoglobin (M 2+ The mass-to-charge accuracy, stability, resolution and repeatability of the instrument were calibrated using a mixed protein standard of 1,4-dihydropyrimidine (DTP) (8476 m / z) and cytochrome C (12355 m / z).

[0063] 2.3 Data Analysis:

[0064] 2.3.1MS file processing

[0065] The peak profile data were detected by mass spectrometry, and the threshold was set to P<0.01 and |log2(FC)|>1 to screen the differential mass spectrometry peaks. The LASSO dimension reduction was performed on the differential mass spectrometry peaks to screen the mass spectrometry characteristic peaks. The ROC method was used to analyze the identification efficiency of mass spectrometry characteristic peaks for ankylosing spondylitis.

[0066] 2.3.2 Artificial neural network algorithm modeling:

[0067] The artificial neural network algorithm was used to model the mass spectrometry characteristic peaks using the neuralnet package in R language, and the ROC curve was drawn.

[0068] 3 Results

[0069] 3.1 Basic characteristics of the subjects:

[0070] A total of 16 healthy individuals and 32 AS patients were included in the study. HLA-B27 tests were all negative in the healthy individuals group. HLA-B27 tests were all positive in the AS group. The basic information of the individuals included in the study is shown in Table 1.

[0071] Table 1

[0072] Grouping Healthy group (n=16) AS group (n=32) age 41.82±15.52 45.86±16.90 Gender (M / F) 8 / 8 17 / 15 HLA-B27 Negative Positive

[0073] 3.2 Qualitative analysis:

[0074] Plasma samples were tested by MALDI-TOF MS, and 188,363 profile peak data were obtained for each sample. Figure 1 Based on the above profile peak data, PCA and UMAP dimensionality reduction analysis showed that there were differences between the two groups.

[0075] 3.3 There are significant differences in mass spectrometry peaks between groups:

[0076] Differential mass peaks (DMPs) were defined as mass peaks with significant differences between the two groups. A total of 3234 DMPs were found between the two groups, of which 105 were upregulated and 3129 were downregulated. Figure 2 and Figure 3 The calculation formula is as follows:

[0077] DMP=p<0.01∩|log2FC|>1

[0078] To further screen the mass spectrometry characteristic peaks, the LASSO regression algorithm was used. Figure 4 As shown, the mass spectrum characteristic peaks are screened, and the calculation process is shown as follows:

[0079]

[0080] Where J(θ) is the objective function of Lasso regression; MSE(θ) is the mean squared error; α is the regularization parameter, by controlling a, the overall regression coefficient can be compressed; θ1, θ2, ···θ n are the parameters of the model.

[0081] LASSO regression screening obtained 16 DMPs with mass-to-charge ratios (M / Z) of 92.2216, 178.5946, 207.7411, 208.4742, 342.9619, 386.2689, 1397.8525, 3152.613, 3154.0399, 8356.3611, 8545.0171, 8565.3403, 8609.5261, 8714.1381, 9534.0368 and 10165.3102 as mass spectrometry characteristic peaks. Figure 5 As shown in Table 2, the identification efficiency of 16 mass spectral characteristic peaks for AS is shown in Table 2:

[0082] Table 2

[0083] M / Z(Da) Specificity(%) Sensitivity(%) AUC (%) 95% CI 92.2216 87.5 84.4 92.8 85.8-99.8 178.5946 62.5 100.0 84.2 70.4-98.0 207.7411 100.0 87.5 94.7 88.7-100.0 208.4742 100.0 87.5 95.1 89.0-100.0 342.9619 100.0 68.8 87.9 78.4-97.3 386.2689 87.5 87.5 89.5 80.4-98.5 1397.8525 81.2 90.6 88.1 76.4-99.8 3152.613 68.8 96.9 81.1 64.5-97.7 3154.0399 68.8 100.0 85.0 71.0-98.9 8356.3611 93.8 62.5 82.4 70.1-94.8 8545.0171 87.5 87.5 87.9 76.6-99.1 8565.3403 81.2 87.5 86.4 74.7-98.2 8609.5261 93.8 84.4 92.4 85.2-99.6 8714.1381 62.5 100.0 89.3 79.8-98.7 9534.0368 62.5 100.0 88.3 78.4-98.2 10165.3102 68.8 96.9 82.6 70.1-95.1

[0084] 3.4 Artificial neural network modeling based on mass spectrometry characteristic peaks:

[0085] The data were divided into a training set (healthy group: 10 cases, AS group: 20 cases) and a test set (healthy group: 6 cases, AS group: 12 cases). The artificial neural network algorithm (including 2 hidden layers, 4 (2×2) ganglia) was used to iterate the training set data for 220 times, and an artificial neural network model based on mass spectrometry characteristic peaks was established. Figure 6 The calculation process is shown in the following formula:

[0086] A. Error term of the output layer

[0087]

[0088] in, is the output of the jth neuron in the output layer, y j is the target value and σ' is the derivative of the activation function.

[0089] B. Error term of hidden layer

[0090]

[0091] in, is the derivative of the activation function, is the weight from the jth neuron in layer I to the kth neuron in layer I+1, is the error term of the I+1th layer, is the output of the kth neuron in layer I.

[0092] C. Gradient of loss function with respect to bias

[0093]

[0094] in, is the error term of the jth neuron in layer I.

[0095] D. Gradient of loss function with respect to weights

[0096]

[0097] in, is the input of the kth neuron in layer I.

[0098] Use the artificial neural network model to analyze and predict the data of the test set, such as Figure 7 The results showed that the prediction accuracy was 100.0%, the sensitivity was 100.0%, the specificity was 100.0%, the positive predictive value was 100.0%, and the negative predictive value was 100.0%. Compared with the single mass spectrometry characteristic peak, the efficiency of the artificial neural network model was significantly improved.

[0099] The above research results show that the neural network model based on plasma molecular fingerprint has excellent identification efficiency for ankylosing spondylitis, and provides new ideas and methods for the identification and prognosis evaluation of ankylosing spondylitis.

[0100] HLA-B27 is a type I MHC gene, which is expressed on all nucleated cells in the body, especially on the surface of lymphocytes. Many years ago, people have found that the expression of HLA-B27 antigen is highly correlated with ankylosing spondylitis. More than 90% of patients with ankylosing spondylitis have positive HLA-B27 anti-inflammatory expression, while only 5%-10% of the general population are positive. Therefore, HLA-B27 detection is of great significance in ankylosing spondylitis. The inventors found that the erythrocyte sedimentation rate increased during the activity of ankylosing spondylitis and the level of inflammatory factors changed significantly. Furthermore, the inventors found that the plasma molecular fingerprint of AS patients was significantly different from that of the normal population through MALDI-TOF MS technology, and the mass spectrometry characteristic peaks were obtained through LASSO regression screening, and a prediction model for AS was developed in combination with a machine learning algorithm. Compared with the traditional HLA-B27 test that only provides a single negative or positive result, the plasma molecular fingerprint prediction model based on machine learning can provide a more detailed and comprehensive biomarker analysis, reflecting the overall metabolic state and pathological process of the patient.

[0101] Those skilled in the art can understand that the above are only preferred examples of the invention and are not intended to limit the invention. Although the invention is described in detail with reference to the above examples, those skilled in the art can still modify the technical solutions recorded in the above examples or replace some of the technical features with equivalents. Any modification, equivalent replacement, etc. made within the spirit and principle of the invention should be included in the protection scope of the invention. All technical features in this embodiment can be freely combined according to actual needs.

[0102] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing a prediction model for ankylosing spondylitis based on a neural network, characterized in that: The specific steps include: Step 1, specimen collection and processing preparation; Step 2, mass spectrometry identification: mass spectrometry analysis is performed using the linear mode of the mass spectrometer; Step 3, data analysis: perform LASSO dimensionality reduction on the differential mass spectrum peaks to screen the mass spectrum characteristic peaks; Step 4: Artificial neural network modeling based on mass spectrum characteristic peaks: The calculation process is shown in the following formula: A. Error term of the output layer in, is the output of the jth neuron in the output layer, y j is the target value, σ' is the derivative of the activation function; is the weighted input of the activation function of the jth neuron in the Lth layer; B. Error term of hidden layer in, is the derivative of the activation function, is the weight from the jth neuron in layer I to the kth neuron in layer I+1, is the error term of the I+1th layer, is the output of the kth neuron in layer I; C. Gradient of loss function with respect to bias in, is the error term of the jth neuron in layer I; D. Gradient of loss function with respect to weights in, is the input of the kth neuron in layer I.

2. The method for constructing a prediction model for ankylosing spondylitis based on a neural network according to claim 1, characterized in that: In step 1, blood samples are collected; the blood samples include blood samples from ankylosing spondylitis patients and healthy people; After sample collection, the samples were centrifuged to separate plasma and stored at -80°C.

3. The method for constructing a prediction model for ankylosing spondylitis based on a neural network according to claim 1, characterized in that: In step 1, blood sample preparation: plasma sample was vortexed with methanol / chloroform mixture at room temperature, centrifuged, and supernatant was collected; the supernatant sample was diluted with sterile deionized water, vortexed and mixed, and the sample was pipetted onto a metal target plate for spotting, and MS-CHCA matrix containing α-cyano-4-hydroxy-cinnamic acid, ethanol and acetonitrile was added; after drying at room temperature, it was tested on the machine.

4. The method for constructing a prediction model for ankylosing spondylitis based on a neural network according to claim 1, characterized in that: In step 2, the linear mode of the Axima Assurance mass spectrometer of Shimadzu Corporation of Japan was used for mass spectrometry analysis; the parameters were as follows: detection method: positive ion; precursor ion scanning range: 1-20000 m / z; referring to "JJF1528-2015 Calibration Specification for Time-of-Flight Mass Spectrometer" and "YYT 1740.2-2021 Medical Mass Spectrometer Part 2: Matrix-Assisted Laser Desorption Ionization Time-of-Flight Mass Spectrometer", trypsin inhibitor 6512 m / z, myohemoglobin M 2+ The mass-to-charge accuracy, stability, resolution and repeatability of the instrument were calibrated using a mixed protein standard substance of 8476 m / z and cytochrome C12355 m / z.

5. The method for constructing a prediction model for ankylosing spondylitis based on a neural network according to claim 1, characterized in that: In step 3, Basic characteristics of the subjects: healthy individuals and patients with ankylosing spondylitis were included; the HLA-B27 test results of the healthy individuals were all negative; the HLA-B27 test results of the ankylosing spondylitis patients were all positive; Qualitative analysis: Plasma samples were tested by MALDI-TOF MS, and 188,363 profile peak data were obtained for each sample. PCA and UMAP dimensionality reduction analysis based on profile peak data showed differences between the two groups. There were significant differences in mass spectrometry peaks between the two groups: the difference mass spectrometry peak DMP was defined as the mass spectrometry peak with significant difference between the two groups; a total of 3234 DMPs were found between the two groups, of which 105 were upregulated and 3129 were downregulated; the calculation formula is as follows: DMP=p<0.01∩|log2FC|>1 Where p is the significant statistic of the comparison of the mass spectrometry peak values ​​between the disease group and the control group, and FC is the ratio of the mean mass spectrometry peak values ​​between the disease group and the control group; In order to further screen the mass spectrometry characteristic peaks, the mass spectrometry characteristic peaks were screened by LASSO regression algorithm. The calculation process is shown in the following formula: Where J(θ) is the objective function of Lasso regression; MSE(θ) is the mean squared error; α is the regularization parameter, by controlling a, the overall regression coefficient can be compressed; θ1, θ2, ...θ n are the parameters of the model; LASSO regression screening obtained 16 DMPs with mass-to-charge ratios of M / Z 92.2216, 178.5946, 207.7411, 208.4742, 342.9619, 386.2689, 1397.8525, 3152.613, 3154.0399, 8356.3611, 8545.0171, 8565.3403, 8609.5261, 8714.1381, 9534.0368 and 10165.3102 as mass spectral characteristic peaks.

Images