Protein marker related to premature delivery, method for constructing premature delivery prediction model, and use thereof

By discovering and using protein markers related to premature birth in cervical vaginal secretions, a premature birth risk assessment model was constructed, which solved the problem of insufficient sensitivity and specificity of premature birth prediction models in the prior art, and achieved higher prediction accuracy and clinical Application value.

WO2024227356A1PCT designated stage expired Publication Date: 2024-11-07BGI GENOMICS CO LTD +1

Patent Information

Application Number
PCT/CN2023/142869
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-12-28
Publication Date
2024-11-07

AI Technical Summary

Technical Problem

The existing predictive model of premature birth is insufficient in clinical applications, resulting in high false positive rates and inability to effectively predict pregnant women with high risk of premature birth, affecting the effectiveness of intervention measures.

Method used

Through the study of proteomic data in cervical vaginal secretions, it was found that protein markers such as basement glycan, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 and other combinations were established to construct premature birth Risk assessment prediction model, using these markers for non-invasive detection and prediction.

Benefits of technology

The accuracy of predicting premature birth risk is significantly improved. The internal verification set AUC is 0.95, and the external verification set AUC is 0.90 and 0.92, respectively, with a specificity of 96.4%, sensitivity of 76.2%, negative predictive value of 94.7%, and positive predictive value of 80.0%, which can be the earliest. Premature birth risk is predicted 9 weeks ahead of the 25th week of pregnancy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2023142869_07112024_PF_FP_ABST
    Figure CN2023142869_07112024_PF_FP_ABST
Patent Text Reader

Abstract

Provided in the present invention are protein markers related to premature delivery, a method for constructing a premature delivery risk assessment and prediction model, a prediction model, a prediction system and the related use thereof. The present invention discovers several protein markers related to premature delivery for the first time, and applies same to the field of premature delivery prediction for the first time. The protein markers related to premature delivery comprise: lumican, β2 microglobulin and α1 microglobulin. Furthermore, the present invention also discovers the use of several combinations of LUM, B2M, AMBP, metalloproteinase inhibitor 1 (TIMP1) and fibronectin 1 (FN1) in the prediction of premature delivery. The premature delivery risk prediction model provided in the present invention, which is constructed by means of using the protein markers in 426 samples has a high accuracy, and has an important value for risk prediction and the monitoring of premature delivery.
Need to check novelty before this filing date? Find Prior Art

Description

Protein markers associated with premature birth, construction methods of premature birth prediction models and their applications Technical Field

[0001] The present invention relates to the field of biomedicine, and in particular to protein markers associated with premature birth, methods for constructing a premature birth risk assessment prediction model, a prediction model, a prediction system, and related applications. Background Art

[0002] Premature delivery (PTB) generally refers to childbirth between 28 and 37 weeks of gestation. Newborns born at this time are called premature babies. Existing data show that the number of premature babies worldwide has reached and exceeded 1 / 10 of the total number of newborns. At the same time, the younger the gestational age, the lower the survival rate of premature babies. Premature babies have an increased risk of illness due to immature organ systems and low resistance. Premature birth is also the leading cause of death in children under 5 years old. Therefore, researchers have been committed to actively studying premature birth prediction methods during pregnancy and developing effective premature birth prediction products to intervene in premature birth outcomes.

[0003] In 1995, the U.S. Food and Drug Administration (FDA) approved the Rapid fFN Test, a test kit developed by Hologic, Inc., for predicting the risk of preterm birth. This patent (EP1936376B1) describes the use of fetal fibronectin (fFN) protein levels in maternal body fluids (urine, whole blood, plasma, saliva, or cervical or vaginal secretions) between 20 and 36 weeks of gestation to predict the risk of preterm birth. Research results show that this method, which measures fFN, has a sensitivity of 83.3% and a specificity of 84.2% for predicting delivery within 14 days in patients at high risk of preterm birth, with negative and positive predictive values ​​of 99.4% and 14.6%, respectively. While this method has an excellent negative predictive value, meaning it accurately predicts women who will not deliver preterm, only 14.6% of women predicted to deliver preterm, resulting in a high false-positive rate and limited clinical utility.

[0004] In 1999, OY MEDIX BIOCHEMICA AB launched the Actim PROM Test, a test kit used to predict the risk of preterm birth or determine imminent delivery. A negative test result clearly indicates that the patient will not deliver within 7-14 days. This patent (WO1999058974A1) assesses cervical ripening and, therefore, the risk of preterm delivery by detecting phosphorylated insulin-like growth factor binding protein 1 (phIGFBP-1) in cervicovaginal secretions. Research results show that this method, which measures phIGFBP1, has a sensitivity of 71.8% and a specificity of 75.3% for predicting delivery within 14 days in patients at high risk of preterm birth, with negative and positive predictive values ​​of 95.9% and 24.8%, respectively. This method also performs well in terms of negative predictive value, with an accuracy of 95.9% for predicting term delivery. However, only 24.8% of women predicted to deliver preterm actually deliver, resulting in a high false-positive rate and limited clinical utility.

[0005] In 2021, Sera Prognostics, Inc. launched the PreTRM Test, which is used to detect two proteins in the blood of pregnant women: insulin-like growth factor-binding protein 4 (IBP4) and sex hormone-binding globulin (SHBG) to predict the gestational age of delivery and the risk of premature birth. This patent (CN108450003B) predicts the gestational age of delivery and the risk of premature birth in pregnant women by the ratio of the two proteins, IBP4 / SHBG. Specifically, the technology uses two protein markers in pregnant women's blood samples to construct a prediction model through the ratio of the two (IBP4 / SHBG), but the performance of the constructed model is AUC of 0.75 in the validation set, and the prediction performance is poor. In addition, the prediction model needs to limit the prediction objects of pregnant women with a BMI between 22-37 and a gestational age of 19-21 weeks. Therefore, this technical method needs to further improve its predictive performance before it can be used in clinical practice.

[0006] In 2021, Changsha Fuyin Biotechnology Co., Ltd. analyzed the combined expression spectrum of any one or at least two proteins in maternal body fluid samples (blood, plasma, serum, urine) to predict whether the subject is premature based on the expression spectrum. This patent (CN202110804975) detects at least one of the IL-1ra protein, TR protein, CCL3 protein, and FS protein in the body fluid, calculates the probability value through the prediction model according to the concentration of the biomarker, and compares the probability value with the size of the model threshold to distinguish people at high risk of premature birth. However, in the embodiment of this application, only IL-1ra protein and TR protein in body fluids were measured, and the types of proteins covered were relatively small. The AUC of its model in the test set was 0.849, but its sensitivity and specificity were not reflected in the study. The study only showed its predictive performance in the training set, with a sensitivity and specificity of 78.8% and 84.3%. This technology does not provide negative predictive values ​​and positive predictive values. Current research data show that this method has poor sensitivity and specificity for screening high-risk premature births.

[0007] In 2018, the U.S. Food and Drug Administration (FDA) approved Qiagen's PartoSure Test kit for predicting delivery time or preterm birth, or for determining the risk of spontaneous chorionic villus rupture in pregnant women. The patent (CN07727841B) describes the detection of placental alpha microglobulin-1 (PAMG-1) concentrations in vaginal secretions from pregnant women aged 20 to 36 weeks and 6 days. A level above a predetermined threshold of 4ng / ml indicates a risk of delivery within the short term (48 hours, 7 days, and / or 14 days). For the PAMG1 marker measured by the technique, the method demonstrated the best predictive performance of all currently available models. Research results demonstrated that the method had a sensitivity of 53.9% and a specificity of 97.4% for predicting delivery within 14 days in patients at high risk of preterm birth, with negative and positive predictive values ​​of 94.9% and 70%. This method's negative predictive value decreased compared to the previous two techniques, but its positive predictive value was significantly improved. Among women predicted to deliver prematurely, 70% ultimately delivered prematurely, significantly reducing false positives. While this method's model predictive specificity significantly improved compared to the aforementioned methods, its sensitivity remained suboptimal. Therefore, this method's clinical value is limited.

[0008] The clinical value of a model is generally evaluated using negative and positive predictive values ​​(PPVs). However, these values ​​are influenced by the proportion of positive cases in the sample. Generally, holding the model's performance (sensitivity / specificity) constant, higher prevalence in the study cohort indicates a higher PPV (Figure 1). Therefore, when comparing the predictive ability of models across studies, sensitivity and specificity are often used to measure model performance. Due to the unique nature of preterm birth interventions, subsequent interventions such as progesterone, cervical cerclage, magnesium sulfate, or glucocorticoids can pose certain risks to maternal health. Therefore, when applying preterm birth prediction models clinically, it is important to ensure a specificity greater than 90% and a minimum false positive rate. Furthermore, sensitivity should be maximized to ensure that more positive cases are predicted. Currently, the best predictive model (the PartoSure Test) has a specificity of 97.4%, a low false positive rate of 2.6%, and a sensitivity of 53.9%, thus only predicting approximately half of positive cases. Therefore, although existing technologies have conducted research and exploration in many fields on premature birth, the actual application of current research results is still insufficient, and the clinical prediction performance still needs to be improved.

[0009] Summary of the Invention

[0010] Through an in-depth and systematic study of the proteomic data of cervicovaginal secretion samples from pregnant women with spontaneous preterm birth and full-term birth, this paper discovered several protein markers with predictive value for preterm birth and applied them to the prediction of preterm birth risk. It also developed a new method for constructing a preterm birth risk assessment prediction model and its related applications.

[0011] In the first aspect, the present invention has discovered for the first time several protein markers associated with premature birth and applied them to the field of premature birth prediction for the first time. These protein markers associated with premature birth include: Lumican (LUM), β-2-microglobulin (B2M) and α1-microglobulin (AMBP). At the same time, the present invention also discovered several combinations of LUM, B2M, AMBP, metalloproteinase inhibitor 1 (TIMP1) and fibronectin 1 (FN1), such as a combination of at least two of the above protein markers, for use in premature birth prediction. Specifically, the present invention provides the application of the above protein markers and their different combinations in constructing a premature birth risk assessment prediction model.

[0012] In a second aspect, the present invention provides a detection kit for protein markers related to premature birth, wherein the protein markers detected by the kit are at least one of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1.

[0013] In the third aspect, the present invention provides a method for constructing a premature birth risk prediction model based on the protein markers in the first aspect, that is, using data of at least one protein marker among basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of pregnant women as modeling factors to construct a prediction model, and the pregnant women participating in the model construction include pregnant women who have premature births and pregnant women who have full-term births.

[0014] In a fourth aspect, based on the prediction model obtained in the third aspect, the present invention provides a method for assessing and predicting the risk of premature birth, the method comprising:

[0015] 1) obtaining data on at least one protein marker selected from the group consisting of basement membrane glycan, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1, and fibronectin 1 of the pregnant woman to be tested;

[0016] 2) Inputting the obtained data of at least one protein marker of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of the above-mentioned pregnant woman into the prediction model for processing to obtain the risk value of premature birth of the above-mentioned pregnant woman. When the risk value is higher than the set threshold, the pregnant woman is determined to be at high risk of premature birth.

[0017] In a specific embodiment of the present invention, data on basement membrane glycan, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1, and fibronectin 1 are obtained from a pregnant woman to be tested. The obtained data on basement membrane glycan, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1, and fibronectin 1 are input into a prediction model for processing to obtain a risk value for premature birth for the pregnant woman to be tested. The prediction model parameters are related to factors such as the model construction method and the protein quantitative detection method (platform).

[0018] In a fifth aspect, based on the prediction method of the fourth aspect, the present invention provides a prediction system for premature birth risk assessment, the system comprising:

[0019] 1) a device for acquiring data on at least one protein marker selected from the group consisting of basement membrane glycan, β2 microglobulin, α1 microglobulin, inhibitor of metalloproteinase 1, and fibronectin 1 of a pregnant woman to be tested;

[0020] 2) a device for performing prediction model processing on the data of at least one protein marker of basement membrane glycan, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1, and fibronectin 1 of the pregnant woman obtained in 1);

[0021] 3) A device for outputting prediction results.

[0022] In a sixth aspect, the present invention provides a product for predicting and assessing the risk of premature birth, comprising: a memory and a processor. The memory is configured to store a program; the processor is configured to execute the program stored in the memory to implement the method for predicting and assessing the risk of premature birth as described in the fifth aspect.

[0023] At the same time, the present invention also provides a computer-readable storage medium, on which a program is stored, and the program can be executed by a processor to implement the premature birth risk assessment and prediction method mentioned in the fourth aspect above.

[0024] In addition, the present invention also provides a computer-readable storage medium, which stores the prediction model obtained by the construction method of the third aspect.

[0025] The beneficial effects of the present invention are:

[0026] This study, for the first time, discovered multiple protein markers associated with preterm birth, providing new detection targets for more accurate prediction of preterm birth risk. Based on the screened protein markers and their combinations, the present invention constructed a preterm birth risk prediction model from 426 samples, demonstrating significantly higher accuracy than existing technologies. Its internal validation set AUC was 0.95, while the two external validation sets were 0.90 and 0.92, respectively. The prediction had a specificity of 96.4±0.0%, a sensitivity of 76.2±7.8%, a negative predictive value of 94.7±2.3%, and a positive predictive value of 80.0%±12.7%.

[0027] In addition, the prediction model provided by the present invention can predict the risk of premature birth as early as 25 weeks of pregnancy and up to 9 weeks in advance, and the risk of premature birth can be predicted in advance using a non-invasive method by simply collecting cervical and vaginal secretions from pregnant women. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic diagram showing the relationship between the prevalence and positive predictive value of the study cohort when the performance (sensitivity / specificity) of the prediction model itself remains unchanged;

[0029] FIG2 is a trend data graph for determining the number of proteins required for the model in Example 1 of the present invention, wherein FIGA represents the median value of the performance (AUC) of all models at a specific number of proteins, and FIGB represents the maximum value of the performance (AUC) of all models at a specific number of proteins;

[0030] FIG3 shows the performance data of the prediction model constructed in Example 1 of the present invention on the internal validation set and the external validation set;

[0031] FIG4 is a schematic diagram showing the results of the ELISA and fluorescence immunochromatography methods for verifying the trend control of five markers in Example 1 of the present invention, wherein PTB represents premature birth and Ctr represents a normal control. DETAILED DESCRIPTION

[0032] Based on the limitations of the existing technology, the present invention uses the differences in protein expression levels in cervical and vaginal secretions to find risk prediction protein markers before the occurrence of premature birth diseases, especially for pregnant women at high risk of premature birth in the second and late stages of pregnancy (with a history of premature birth, multiple miscarriages, cervical dilation ≤3cm or premature birth symptoms such as uterine contractions, intermittent lower abdominal pain and pelvic pressure, and meeting one of the above conditions), using a non-invasive method to predict the risk of premature birth diseases in advance and construct a premature birth risk prediction model.

[0033] First, the present invention systematically analyzed the expression profiles of proteins in maternal cervicovaginal secretions during pregnancy, compared the differential characteristics of the protein expression profiles of cervicovaginal secretions in pregnant women with spontaneous preterm birth and pregnant women with full-term delivery, and screened out protein markers related to the prediction of premature birth from a wide variety of proteins, laying the foundation for the subsequent construction of a more accurate premature birth risk prediction model. At the same time, by analyzing and studying the proteins in cervicovaginal secretions, the value of different types of protein molecules in cervicovaginal secretions in predicting premature birth was also revealed. Among them, LUM, B2M and AMBP were found to be associated with premature birth for the first time by the present invention, and can be used as protein markers for premature birth, which has not been mentioned in previous reports. At the same time, the present invention also proposed for the first time the application of a combination of one or more of LUM, B2M, AMBP, TIMP1 and FN1 in the field of premature birth prediction. The above protein markers and their clinical significance are shown in Table 1.

[0034] Table 1 Protein markers and their clinical significance

[0035] The above-mentioned protein markers LUM, B2M, AMBP, TIMP1, and FN1 can be detected using a variety of methods, such as protein arrays, proteomics, expression proteomics, mass spectrometry (e.g., liquid chromatography-mass spectrometry (LC-MS), multiple reaction monitoring (MRM), selected reaction monitoring (SRM), scheduled MRM, scheduled SRM, 2D PAGE, 3D PAGE, electrophoresis, proteome chips, proteome microarrays, Edman degradation, direct or indirect ELISA, immunosorbent assay, immuno-PCR, proximity extension assay, Luminex assay or homogenization assay, time-resolved fluorescence (TRF), fluorescent oxygen channeling immunoassay (FOCI) or luminescent oxygen channeling immunoassay), targeted protein detection techniques using liquid chromatography-mass spectrometry, chemiluminescent immunoassay, and fluorescent immunoassay, all of which can produce similar detection results.

[0036] In addition, in addition to the above-mentioned protein markers themselves being used as detection objects in the prediction of premature birth, the expression levels or transcriptome information of the genes corresponding to these proteins can also be detected and applied to the prediction of premature birth. Therefore, the application of the above-mentioned protein markers in the prediction of premature birth involves multiple aspects and different forms. For example, such an application can be a kit for accurately detecting the above-mentioned protein markers, or a related reagent for analyzing the transcriptome of the above-mentioned protein markers, etc. At the same time, it can also be to combine these detection or analysis results and use machine learning to construct a premature birth risk prediction model and perform risk prediction. The gene information of the above-mentioned protein markers is shown in Table 2.

[0037] Table 2 Biological information of protein markers

[0038] Regarding the detection of the aforementioned protein markers, in one embodiment of the present invention, protein extraction is performed on cervicovaginal secretions obtained from pregnant women. It should be understood that in addition to cervicovaginal secretions, extraction, detection, and analysis of relevant protein markers or their transcriptomes can also be performed on other sample types, such as whole blood, plasma, amniotic fluid, serum, and urine.

[0039] Secondly, the present invention utilizes the above-mentioned protein markers and their different combinations to construct a model for predicting the risk of premature birth. The construction method includes: obtaining relevant data of at least one protein marker among basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of pregnant women as relevant modeling factors, and constructing a training sample set with these modeling factors and their corresponding pregnant women, and the above-mentioned pregnant women include pregnant women with premature birth and pregnant women with full-term birth. Based on the above-mentioned training sample set, several types of model training are performed, and the several types of models after training are verified and evaluated, and the best prediction model is confirmed according to the model evaluation indicators. It should be understood that the above-mentioned prediction model can usually be evaluated by multiple indicators such as accuracy, precision, ROC curve, AUC, etc., and the best prediction model is confirmed according to the aforementioned indicators.

[0040] In a specific embodiment of the present invention, the above-mentioned construction method includes:

[0041] 1) Determine the model training set samples and internal validation set samples, both of which include pregnant women with preterm births and pregnant women with full-term births;

[0042] 2) using at least one protein marker of basement membrane glycan, β2-microglobulin, α1-microglobulin, metalloproteinase inhibitor 1, and fibronectin 1 from pregnant women in the above model training set as features, a prediction model was constructed using the logistic regression method. The model construction used a 5-fold cross-validation grid search method to automatically adjust the model parameters to obtain the optimized prediction model;

[0043] 3) Use the above internal validation set samples to evaluate the performance of the prediction model in 2).

[0044] After obtaining the above prediction model, we can further use independent verification samples to independently evaluate the performance of the above optimal prediction model.

[0045] Those skilled in the art will appreciate that, in addition to constructing a predictive model using the aforementioned logistic regression method, alternative models can be constructed using data from at least one of the maternal LUM, B2M, AMBP, TIMP1, and FN1 as input, with the likelihood of preterm birth as output, through machine learning, deep learning, or artificial intelligence models to generate a predictive model for preterm birth risk assessment. The algorithm or model type used to construct the predictive model does not constitute a limitation of the present invention. In fact, the aforementioned protein markers and their combinations in the present invention have demonstrated excellent predictive effectiveness in models constructed using a variety of different algorithms.

[0046] In a specific embodiment of the present invention, the above-mentioned construction method includes:

[0047] 1) Determine the model training set samples and internal validation set samples, both of which include pregnant women with preterm births and pregnant women with full-term births;

[0048] 2) using at least one protein marker of basement membrane glycan, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1, and fibronectin 1 from the pregnant women in the above model training set samples as features, and constructing a prediction model using a random forest algorithm;

[0049] 3) Use the above internal validation set samples to evaluate the performance of the prediction model in 2).

[0050] After obtaining the above prediction model, we can further use independent verification samples to independently evaluate the performance of the above prediction model.

[0051] In a specific embodiment of the present invention, the above-mentioned construction method includes:

[0052] 1) Determine the model training set samples and internal validation set samples, both of which include pregnant women with preterm births and pregnant women with full-term births;

[0053] 2) using at least one protein marker of basement membrane glycan, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1, and fibronectin 1 from pregnant women in the above model training set samples as features, and constructing a prediction model using a neural network algorithm;

[0054] 3) Use the above internal validation set samples to evaluate the performance of the prediction model in 2).

[0055] After obtaining the above prediction model, we can further use independent verification samples to independently evaluate the performance of the above prediction model.

[0056] It should be noted that the above description mentioned that LUM, B2M, AMBP, TIMP1 and FN1 and their corresponding various combinations can be fully utilized in model construction to carry out specific model construction, so that multiple groups of different prediction models can be obtained, and multiple groups of different ROC curves (Receiver operating characteristic curve, ROC) can also be obtained. According to different ROC curves, the sensitivity, specificity, positive predictive value and negative predictive value of each model for predicting preterm birth can be calculated respectively, and the optimal model can be selected.

[0057] In one embodiment of the present invention, the threshold is the threshold at which the Youden Index (Youden Index = sensitivity + specificity) is maximized. It should be understood that other methods—model evaluation indicators of the prediction model—can also be used to determine the threshold, such as setting the threshold based on a specific model specificity.

[0058] It should be understood that after the model is constructed by the above method, when performing premature birth risk assessment and prediction, the combination of protein markers used by the corresponding model inputs the LUM, B2M, AMBP, TIMP1 and FN1 data of the pregnant woman to be tested into the prediction model, and the model can automatically calculate the risk value, and evaluate the premature birth risk as high risk, medium risk and low risk according to the risk value.

[0059] The present invention will be further described in detail below through specific embodiments with reference to the accompanying drawings.

[0060] Example 1: Screening of protein markers and construction of a premature birth risk prediction model

[0061] 1. Sample acquisition

[0062] (1) Obtaining maternal cervicovaginal secretion samples

[0063] Cervicovaginal secretions were obtained from pregnant women at the hospital, including those who had both premature and full-term deliveries. Sampling method: Obstetricians used endoscopy to obtain samples. During the speculum examination, a sterile swab was placed in the posterior fornix of the vagina and allowed to soak for 30 seconds. When removing the swab, avoid touching other parts of the vagina. The swab was stored in a dry, sterile cryovial with the cap securely screwed on. It was temporarily stored at 4°C and transferred to -80°C for long-term storage within 8 hours. It was then transported on dry ice.

[0064] The above samples include a training set and an internal validation set (a total of 165 cases, 54 premature births and 111 full-term births). Specific information such as sample demographics and clinical characteristics is shown in Table 3.

[0065] Table 3 Demographic and clinical characteristics of the training set and internal validation set samples

[0066] (2) Extraction of protein from maternal cervicovaginal secretion samples

[0067] a. Break off the swab head and place it into a 2ml EP tube;

[0068] b. Add 500 μl of 1× phosphate buffer (PBS: 10 mmol / L Phosphate buffer; 137 mmol / L NaCl; 2.7 mmol / L KCl; pH 7.4) to the EP tube to wet the swab tip.

[0069] c. The vaginal secretion sample was resuspended in 1 ml of lysis buffer (1% SDS, 100 mM ABC pH 8.5) for lysis, and then sonicated on ice for 20 minutes;

[0070] d. The lysate was centrifuged and the supernatant was reduced and alkylated;

[0071] e. Take 200 μl of protein solution and add it to a 96-well plate containing 20 μl of Magnetic Sera-Mag Speed ​​Beads (containing 10 μl each of magnetic beads A and magnetic beads B). Add 200 μl of 100% acetonitrile (ACN) to each well and incubate for 20 minutes.

[0072] f. Place on a magnetic stand for 5 minutes, remove the supernatant, and wash the beads twice with 200 μl 80% ethanol (EtOH) and once with 200 μl 100% ACN.

[0073] g. Resuspend the beads in 50 μl of 100 mM ABC and sonicate in a water bath for 5 minutes. Add sequencing-grade trypsin (5 μl of 0.1 μg / μl trypsin dissolved in ddH2O) at a 1:20 enzyme:protein ratio and disperse the beads to ensure efficient digestion.

[0074] h. The samples were incubated overnight in a thermomixer at 37°C and 1000 rpm. The peptides bound to the beads were eluted with 2% ACN containing 0.1% FA, and the supernatant was collected in a new tube for LC-MS analysis.

[0075] 2. Acquisition of proteomic data

[0076] a. Using a capture column (Acclaim PepMap C18, 5 μm, The digested peptides were separated by an Ultimate 3000 RSLC (Thermo Fisher Scientific, USA) with a diameter of 100 μm × 2 cm, and then the peptides were analyzed using a homemade C18 analytical column (1.8 μm, The outlet of the analytical column was directly connected to an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Fisher Scientific, USA).

[0077] b. For MS setup, the capillary temperature was set to 325°C. Data-independent acquisition (DIA) mode was used for protein quantification. Full-scan MS spectra were acquired in the Orbitrap over the mass range m / z 400–1,200 at a resolution of 60,000 FWHM.

[0078] c. The raw files were processed using DIA-NN (version 1.8.1). DIA-NN uses a protein sequence database (UniProt, UP000005640, 20431 proteins) to predict a spectral library, which was then used to search and quantify the DIA data. The precursor and protein FDRs were set to 1%.

[0079] 3. Screening of protein markers for the prediction of premature birth risk and construction of prediction models

[0080] 1) Preliminary filtering of mass spectrometry data: proteins with missing values ​​greater than 20% were eliminated.

[0081] 2) Preliminary screening of protein markers for the prediction of preterm birth risk: In a sample of 165 cases (54 preterm and 111 term), 3855 proteins identified by mass spectrometry were preliminarily screened using the Jackknifing and Elastic net (EN) algorithms. A total of 293 proteins were identified by the two methods as potentially predictive of preterm birth. The first method, using the Jackknifing method, identified a total of 264 biomarkers. The Jackknifing method employed 1000 repeated random samplings, each extracting 90% of the preterm and term samples. A statistical analysis of differences was performed, and the results of the 1000 differential analyses were pooled. The upper limit of the confidence interval of the 1000 corrected P values ​​was less than 0.05, and the up- and down-regulation trends of the 1000 identified proteins were the same. Based on these identification criteria, a total of 179 upregulated proteins and 85 downregulated proteins (a total of 264 proteins) were obtained. The second approach employed LOOCV (leave-on-out CV) cross-validation, combined with the EN (Elastic Net) algorithm, to evaluate the performance of the prediction model by screening different numbers of proteins. A total of 2, 3, 4, … 50 protein combinations were screened for predictive performance, and a total of 96 proteins were selected during model construction. Using these two methods, a total of 293 protein markers were initially identified (67 of which were found in both methods). To further narrow the selection, a corrected P value of less than 0.01 and an area under the AUROC curve of greater than 0.7 for each marker were used as thresholds. This resulted in a preliminary selection of 112 proteins as candidate protein markers.

[0082] 3) Protein Marker Combination Screening: To achieve optimal predictive performance using the fewest protein markers, this example further traverses all protein combinations (2, 3, 4…10 proteins) for the 112 candidate protein markers obtained in the previous step. Considering the potential for subsequent clinical translational applications, protein markers with commercially available antibodies (63) were prioritized in the combination process. The training set was divided into two parts: a model construction set and an internal validation set (model construction set: 27 preterm births and 56 term births, internal validation set: 27 preterm births and 55 term births). A predictive model was constructed using logistic regression. A 5-fold cross-validation grid search (GridSearchCV) method was used to automatically adjust the model parameters to obtain the optimal predictive model. Finally, model performance was evaluated in the internal validation set (using the area under the AUROC curve as the model prediction performance metric).

[0083] Analysis revealed that as the number of proteins in the prediction model increased, the model's prediction accuracy gradually increased. As shown in Figure 2, prediction performance gradually increased from one to five proteins, but continued addition of proteins plateaued, with no significant improvement in model performance beyond five proteins. This result suggests that the optimal number of proteins in the model is 4-5, providing a reference for determining the final number of protein combinations.

[0084] 4) Independent Validation of Model Predictive Performance: All samples were used in the above protein marker screening process, so using the internal validation set samples for model evaluation may have data information leakage issues. Therefore, this study used two completely independent external validation set samples (external validation set 1: 11 premature births and 116 full-term births; external validation set 2: 24 premature births and 110 full-term births) to independently evaluate the performance of the above model (a model constructed using any combination of the 63 protein markers with commercially available antibodies selected in 3) above). Ultimately, 42 protein combinations were screened out with AUROC greater than 0.9 in the training set, internal validation set, and external validation set. Specific information such as sample demographics and clinical characteristics of the external validation set samples is shown in Table 4. The relevant sample acquisition and protein extraction and testing were the same as those of the training set and internal validation set.

[0085] Table 4 Demographic and clinical characteristics of the external validation set samples

[0086] 5) Protein marker combination determination: The 42 protein combinations in 4) above were subjected to frequency statistics, and several frequently selected proteins were found. Among them, the top five proteins included LUM, TIMP1, FN1, B2M, and AMBP. These five protein markers were all shown to be upregulated in the differential expression analysis.

[0087] The above five proteins LUM, TIMP1, FN1, B2M and AMBP were used to construct the model using the same prediction model construction method as in the above 3) protein marker combination screening. The prediction performance of the models of different combinations in different data sets is shown in Table 5, where the area under the AUROC curve was used as the model prediction performance indicator.

[0088] Table 5 Prediction performance of various combinations of 5 protein markers

[0089] The prediction model constructed by the logistic regression method using five protein markers at the same time is as follows:

[0090] Risk of premature birth = 1 / (1+ea*AMBP+b*TIMP1+c*B2M+d*FN1+f*LUM-0.85);

[0091] Among them, LUM, B2M, AMBP, TIMP1 and FN1 are the charge-to-mass ratios of LUM, B2M, AMBP, TIMP1 and FN1 obtained by mass spectrometry detection of cervical and vaginal secretion samples of pregnant women, and the coefficients a, b, c, d and f are: a is 5.74e -4 , b is 5.48e -3 , c is 4.15e -3 , d is 1.76e -4 , f is 2.04e -4 .

[0092] The performance data of the above prediction model on the internal validation set and the external validation set are shown in Table 6 and Figure 3.

[0093] Table 6 AUC and specificity of the internal and external validation sets of the optimal model

[0094] 4. Validation of Marker Performance by Enzyme-Linked Immunosorbent Assay: Validation of the five protein markers using ELISA / fluorescent immunochromatography assays was performed. The results were highly consistent with the mass spectrometry proteomics trend control data (as shown in Figure 4), showing that all markers were upregulated. Therefore, ELISA / fluorescent immunochromatography assays can be used for protein marker detection and application in predictive models.

[0095] The above experimental data show that the present invention uses the above five proteins as a feature combination, combined with a machine learning model, to predict the risk of premature birth as early as 25 weeks of gestation and up to 9 weeks in advance. For example, in the external validation of this embodiment, the risk of premature birth can be predicted based on the gestational age of 25 weeks. +6A sample of a pregnant woman enrolled in the group at week 1 was tested for the relative concentrations of the above five proteins. Based on the risk prediction model, the risk value of premature birth for this pregnant woman was output (the model output value was 0.33, which was converted to a relative risk of 85%), indicating a high risk of premature birth. The follow-up outcome was that this pregnant woman was 34 +6 The model accurately predicted the risk of preterm birth in the 426-case external validation set, with an AUC of 0.93 for the training set, 0.95 for the internal validation set, and 0.90 and 0.92 for the external validation set, respectively. The specificity was 96.4±0.0% and the sensitivity was 76.2±7.8%.

[0096] Specifically, the preterm birth prediction model in this example achieved an AUC of 0.95, a specificity of 96.4%, a sensitivity of 85.2%, a negative predictive value (NPV) of 93.1%, and a positive predictive value (PPV) of 92.0% in the first internal validation set of 82 cases (27 preterm births and 55 healthy controls), which are higher than the state of the art. In a second, completely independent external validation set of 127 cases (11 preterm births and 116 healthy controls), the AUC was 0.90, with a specificity of 96.4%, a sensitivity of 72.7%, a negative predictive value (NPV) of 97.3%, and a positive predictive value (PPV) of 66.7%, which are also higher than the state of the art. In a third, completely independent external validation set of 134 cases (24 preterm births and 110 healthy controls), the AUC was 0.92, with a specificity of 96.4%, a sensitivity of 70.8%, a negative predictive value (NPV) of 93.8%, and a positive predictive value (PPV) of 81.0%, which are also higher than the state of the art.

[0097] In addition, those skilled in the art should understand that in addition to the above-mentioned construction of the prediction model using the logistic regression method, at least one of the data of LUM, B2M, AMBP, TIMP1 and FN1 of the pregnant women can be used as input, and whether the pregnant woman has premature birth is used as output. The model can be constructed through machine learning, deep learning or artificial intelligence model substitution to obtain a premature birth risk assessment prediction model, such as the following Examples 2 and 3.

[0098] Example 2

[0099] In this example, the same sample set as in Example 1 (including training set, internal validation set, external validation sets 1 and 2) was used, and the above-mentioned five proteins LUM, TIMP1, FN1, B2M and AMBP were used to construct a prediction model using the random forest algorithm. The prediction performance of the prediction models with different combinations in different data sets is shown in Table 7, where the area under the AUROC curve was used as the prediction performance indicator of the prediction model.

[0100] Table 7 Prediction performance of random forest model for multiple combinations of 5 protein markers

[0101] It can be seen from Table 7 that, similar to the results in Table 5, the AUCs of different combinations of the above five proteins LUM, TIMP1, FN1, B2M and AMBP in the internal validation set and external validation set of the prediction model constructed using the random forest algorithm are generally above 0.9, indicating that the random forest model constructed using different combinations of the five proteins LUM, TIMP1, FN1, B2M and AMBP in the present invention can effectively predict the risk of premature birth.

[0102] Example 3

[0103] In this example, the same sample set as in Example 1 (including training set, internal validation set, external validation sets 1 and 2) was used, and the above-mentioned five proteins LUM, TIMP1, FN1, B2M and AMBP were used to construct a prediction model using a neural network algorithm. The prediction performance of the prediction models with different combinations in different data sets is shown in Table 8, where the area under the AUROC curve was used as the prediction performance indicator of the prediction model.

[0104] Table 8 Prediction performance of neural network models for multiple combinations of 5 protein markers

[0105] It can be seen from Table 8 that, similar to the results in Tables 5 and 7, the different combinations of the above five proteins LUM, TIMP1, FN1, B2M and AMBP have AUCs in the internal validation set and the external validation set of the prediction model constructed using the neural network algorithm that are mostly above 0.9, indicating that the neural network model constructed using different combinations of the five proteins LUM, TIMP1, FN1, B2M and AMBP in the present invention can effectively predict the risk of premature birth.

[0106] The above examples are used to illustrate the present invention, which are only used to help understand the present invention and are not intended to limit the present invention. Those skilled in the art can make several simple deductions, modifications or substitutions based on the concept of the present invention.

Claims

1. Application of at least one of basement membrane polysaccharide, β2-microglobulin and α1-microglobulin as a protein marker in the prediction of premature birth.

2. A composition of protein markers associated with premature birth, It is characterized in that The composition comprises at least two of perlecan, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1.

3. Application of a protein marker combination consisting of at least two of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 in the prediction of premature birth.

4. The application of at least one of the protein markers basement membrane polysaccharide, β2-microglobulin, and α1-microglobulin in constructing a model for predicting the risk of premature birth.

5. The application of at least two of the protein markers basement membrane polysaccharide, β2-microglobulin, α1-microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 in constructing a prediction model for preterm birth risk assessment.

6. A detection kit for protein markers related to premature birth, It is characterized in that The protein marker detected by the kit is at least one of basement membrane polysaccharide, beta 2 microglobulin, alpha 1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1.

7. The detection kit according to claim 6, It is characterized in that The detection method of the protein marker includes: any one of or a combination of ELISA enzyme-linked immunosorbent assay technology, liquid phase mass spectrometer targeted protein detection technology, chemiluminescent immunoassay protein detection technology, and fluorescent immunoassay protein detection technology.

8. The detection kit according to claim 6, It is characterized in that The protein markers are derived from cervicovaginal secretions of pregnant women.

9. A method for constructing a premature birth risk assessment prediction model. It is characterized in that The premature birth risk assessment prediction model is constructed based on data of at least one protein marker among basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of pregnant women as modeling factors, and the pregnant women include pregnant women who have premature birth and pregnant women who have full-term birth.

10. The construction method according to claim 9, It is characterized in that The construction method comprises: Obtaining relevant data of at least one protein marker of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of pregnant women as relevant modeling factors, and constructing a training sample set with the modeling factors and the corresponding pregnant women, wherein the pregnant women include pregnant women with premature delivery and pregnant women with full-term delivery; Several types of model training are performed based on the training sample set, and the trained several types of models are evaluated, and the best prediction model is confirmed according to the model evaluation index.

11. The construction method according to claim 10, It is characterized in that The construction method comprises: 1) Determine the model training set samples and internal validation set samples, which include pregnant women with premature birth and pregnant women with full-term birth; 2) using at least one protein marker of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of pregnant women in the model training set samples as features, using logistic regression method to construct a prediction model, and using 5-fold cross-validation grid search method to automatically adjust the model parameters to obtain the optimal prediction model; 3) Using the internal validation set samples to evaluate the performance of the prediction model in 2).

12. The construction method according to claim 11, It is characterized in that Also included is independent validation: an independent validation sample is used to independently evaluate the performance of the optimal prediction model, and the sample includes pregnant women who have preterm births and pregnant women who have term births.

13. The construction method according to claim 10, It is characterized in that The construction method comprises: 1) Determine the model training set samples and internal validation set samples, which include pregnant women with premature birth and pregnant women with full-term birth; 2) using at least one protein marker of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of pregnant women in the model training set samples as features, and constructing a prediction model using a random forest algorithm; 3) Using the internal validation set samples to evaluate the performance of the prediction model in 2).

14. The construction method according to claim 13, It is characterized in that Also included is independent validation: an independent evaluation of the prediction model performance was performed using an independent validation sample, which included pregnant women who experienced preterm birth and pregnant women who delivered at term.

15. The construction method according to claim 10, It is characterized in that The construction method comprises: 1) Determine the model training set samples and internal validation set samples, which include pregnant women with premature birth and pregnant women with full-term birth; 2) using at least one protein marker of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of pregnant women in the model training set samples as features, and constructing a prediction model using a neural network algorithm; 3) Using the internal validation set samples to evaluate the performance of the prediction model in 2).

16. The construction method according to claim 15, It is characterized in that Also included is independent validation: an independent evaluation of the prediction model performance was performed using an independent validation sample, which included pregnant women who experienced preterm birth and pregnant women who delivered at term.

17. The construction method according to claim 10, It is characterized in that After obtaining the prediction model, the threshold is set according to the model evaluation index of the prediction model.

18. The construction method according to claim 17, It is characterized in that The threshold value is the threshold value when the Youden index is the maximum.

19. A method for assessing and predicting the risk of premature birth. It is characterized in that The prediction method comprises: Acquiring data of at least one protein marker of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of the pregnant woman to be tested; Inputting the obtained data of at least one protein marker of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of the pregnant woman into the prediction model for processing, and obtaining the risk value of premature birth of the pregnant woman to be tested. If the risk value exceeds a threshold value, the pregnant woman to be tested is determined to have a high risk of premature birth; The prediction model is obtained by the construction method according to any one of claims 9 to 18.

20. A prediction system for premature birth risk assessment, It is characterized in that The prediction system comprises: A device for acquiring data of at least one protein marker selected from the group consisting of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1, and fibronectin 1 of a pregnant woman to be tested; A device for performing prediction model processing on data of at least one protein marker of basement membrane polysaccharide, β2 microglobulin, α1 microglobulin, metalloproteinase inhibitor 1 and fibronectin 1 of the pregnant woman; means for outputting a prediction result; The prediction model is obtained by the construction method according to any one of claims 9 to 18.

21. A predictive product for premature birth risk assessment, Features include: Memory, used to store programs; A processor, configured to implement the method according to claim 19 by executing the program stored in the memory.

22. A computer-readable storage medium, It is characterized in that The medium stores a program, which can be executed by a processor to implement the method according to claim 19 .

23. A computer-readable storage medium, It is characterized in that The medium stores a prediction model obtained by the construction method according to any one of claims 9 to 18.

Citation Information

Patent Citations

  • Methods and kits for predicting risk for preterm labor

    CN101509917A

  • Methods for predicting time-to-delivery in pregnant women

    CN105008925A

  • Biomarkers and methods for predicting preterm birth

    CN106574919A

  • Quantitative profiling of progesterone metabolites for the prediction of spontaneous preterm delivery

    CN109415402A

  • Protein molecular marker for premature delivery prediction and prediction method thereof

    CN115616217A

Cited By

  • Multi-liver pathology assessment method and system based on machine learning

    CN119905274A

  • Protein biomarker for auxiliary diagnosis of PCOS spontaneous premature delivery and application thereof

    CN121208353A