A biomarker and risk assessment system for predicting atopic dermatitis

Abstract

The application belongs to the technical field of disease risk prediction, and particularly relates to a biomarker for predicting atopic dermatitis and a risk assessment system. The application establishes a method for predicting the risk of atopic dermatitis in infants by collecting umbilical cord blood serum samples at birth, screening candidate proteins, quantitatively detecting immune-related markers, screening variables, machine learning modeling, external verification and model simplification. The method can be completed in the early infancy of the infant. The application improves the stability and repeatability of the results through a standardized detection process and variable screening rules, and can reduce the application complexity while maintaining high performance through model simplification. The application has good prediction performance and external generalization ability, and is more suitable for clinical popularization and application.

Description

Technical Field

[0001] This invention belongs to the field of disease risk prediction technology, specifically relating to a biomarker and risk assessment system for predicting atopic dermatitis. Background Technology

[0002] Atopic dermatitis (AD) is a common chronic inflammatory skin disease in infancy and early childhood. It usually begins in early life and is closely related to the occurrence of subsequent allergic diseases. Therefore, identifying and intervening in high-risk individuals in the early postnatal period is of great clinical significance.

[0003] As our understanding of the pathogenesis of Alzheimer's disease (AD) deepens, research has shown that the occurrence of AD is closely related to immune imbalance and involves multiple mechanisms such as type II inflammation, chemokine networks, and neuro-immune pathways. Current early risk assessments for Alzheimer's disease (AD) primarily rely on clinical data such as family history of allergies, perinatal factors, feeding methods, and environmental exposures. However, the correlation between the model indicators included in these methods and the pathogenesis of AD is limited, resulting in the following shortcomings: First, they rely mainly on clinical risk factors for risk assessment, leading to limited predictive accuracy and failing to meet the needs of individualized early screening for newborns. Second, they fail to fully utilize the immune-related information contained in umbilical cord blood at birth, making it difficult to identify key high-risk signals before the disease develops. Third, they often use single indicators or a small number of non-specific immune features for analysis, failing to fully reflect the multi-pathway biological basis of AD development, thus limiting predictive ability and interpretability. Fourth, they lack independent external validation, resulting in insufficient model generalization ability and difficulty in directly extending to different populations. Fifth, they do not adequately consider aspects such as standardized testing procedures, processing of data below the detection limit, batch consistency, and data traceability, affecting the stability and repeatability of results. Sixth, some protocols have a large number of features and complex procedures, hindering clinical application. Therefore, constructing a predictive model that can accurately predict the risk of AD and has generalization ability has important clinical value. Summary of the Invention

[0004] The purpose of this invention is to provide a biomarker for predicting atopic dermatitis in infants and young children.

[0005] Another object of the present invention is to provide the application of the above-mentioned biomarkers in the preparation of products for predicting atopic dermatitis in infants and young children.

[0006] Another object of the present invention is to provide a risk assessment system for atopic dermatitis in infants and young children.

[0007] The application of biomarkers according to specific embodiments of the present invention in the preparation of products for predicting the risk of developing atopic dermatitis in subjects, wherein the biomarkers are one or more combinations of BDNF, CTACK / CCL27, Eotaxin / CCL11, Eotaxin-2 / CCL24, Eotaxin-3 / CCL26, GM-CSF / CSF2, I-TAC / CXCL11, IL-2, IL-10, IL-21, IP-10, MDC / CCL22, MIG / CXCL9, MIP-1α / CCL3, MIP-3α / CCL20, RANTES / CCL5, BAT3 / BAG6, TGF-β1 / TGFB1, CCL14, COL6A3, TIMP3, POSTN / Periostin, and CCL17 / TARC.

[0008] In this invention, the application value of the biomarkers lies in improving the accuracy and reliability of diagnosis through the joint analysis of multi-dimensional information.

[0009] According to specific embodiments of the present invention, the product is a reagent kit, chip, reagent, or monitoring system for detecting biomarkers.

[0010] In this invention, the term "biomarker" refers to an indicator that can be detected in a sample, such as a predictive, diagnostic, and / or prognostic indicator. Biomarkers can serve as indicators of a specific disease or condition (e.g., cancer) subtype characterized by specific molecular, pathological, histological, and / or clinical features. In some embodiments, a biomarker is a gene or protein. Biomarkers include, but are not limited to, polynucleotides (e.g., DNA and / or RNA), polynucleotide copy number alterations (e.g., DNA copy number), peptides, peptide and polynucleotide modifications (e.g., post-translational modifications), carbohydrate and / or glycolipid-based molecular markers.

[0011] In this invention, the term "expression level" generally refers to the amount of a biomarker in a biological sample. "Expression" generally refers to the process by which information (e.g., gene coding and / or epigenetics) is transformed into structures present and functioning in the cell. Therefore, "expression" as used in this invention can refer to transcription into polynucleotides, translation into peptides, polynucleotide and / or peptide modifications (e.g., post-translational modifications of peptides). Transcribed polynucleotide fragments, translated peptide fragments, or polynucleotide and / or peptide modified fragments (e.g., post-translational modifications of peptides) should also be considered expressed, regardless of whether they originate from transcripts generated by alternative splicing or degraded transcripts, or from post-translational processing of peptides (e.g., through proteolysis).

[0012] In this invention, the term "reagent kit" refers to a commercially available product that pre-packages and combines one or more reagents required for detection. Reagent kits typically also include instructions for use, standards, quality control materials, and auxiliary components such as buffer solutions and substrates required for the reaction. "Specific detection" refers to the ability of the detection reagent to specifically bind or react with the target biomarker, without producing significant cross-reactions to other non-target components in the sample. Specificity is a key technical indicator for achieving accurate quantitative detection.

[0013] In this invention, the term "chip" refers to a gene chip, also known as a DNA chip or biochip, which involves immobilizing probe molecules on a support and then hybridizing them with labeled sample molecules. The quantity and sequence information of the sample molecules are obtained by detecting the hybridization signal intensity of each probe molecule.

[0014] According to a specific embodiment of the present invention, a kit for predicting the risk of developing atopic dermatitis in a subject is provided. The kit includes reagents for detecting the expression levels of biomarkers, wherein the biomarkers include one or more combinations of BDNF, CTACK / CCL27, Eotaxin / CCL11, Eotaxin-2 / CCL24, Eotaxin-3 / CCL26, GM-CSF / CSF2, I-TAC / CXCL11, IL-2, IL-10, IL-21, IP-10, MDC / CCL22, MIG / CXCL9, MIP-1α / CCL3, MIP-3α / CCL20, RANTES / CCL5, BAT3 / BAG6, TGF-β1 / TGFB1, CCL14, COL6A3, TIMP3, POSTN / Periostin, and CCL17 / TARC.

[0015] The present invention also provides a system for predicting the risk of developing atopic dermatitis in a subject, the system comprising a detection module, a model calculation module, and a risk determination module, wherein, The detection module acquires the following biomarker content data of the subject: BDNF, CTACK / CCL27, Eotaxin / CCL11, Eotaxin-2 / CCL24, Eotaxin-3 / CCL26, GM-CSF / CSF2, I-TAC / CXCL11, IL-2, IL-10, IL-21, IP-10, MDC / CCL22, MIG / CXCL9, MIP-1α / CCL3, MIP-3α / CCL20, RANTES / CCL5, BAT3 / BAG6, TGF-β1 / TGFB1, CCL14, COL6A3, TIMP3, POSTN / Periostin, CCL17 / TARC; The model calculation module, based on the obtained biomarker content data and clinical variables, performs calculations using a random forest model to obtain the risk probability value of the subject's atopic dermatitis. The risk determination module is used to compare the risk probability value with a preset threshold and generate an auxiliary diagnostic conclusion based on the comparison result.

[0016] The "risk assessment system" refers to an auxiliary decision-making tool built based on computer technology, bioinformatics methods, and clinical laboratory data. It can be installed in the computer equipment of the medical laboratory department, or integrated into the control software of Luminex liquid chromatography-chip analyzers and ELISA enzyme-linked immunosorbent assay analyzers, or it can provide online analysis services in the form of a cloud platform.

[0017] According to a specific embodiment of the system of the present invention, the model is constructed based on the random forest algorithm using the content data of the biomarkers and clinical variables as feature information.

[0018] In this invention, the system for predicting the risk of developing atopic dermatitis in a subject refers to an auxiliary decision-making tool built based on computer technology, bioinformatics methods, and clinical laboratory data. It can be installed in the computer equipment of a medical laboratory department, or integrated into the control software of a fully automated chemiluminescence analyzer and a liquid chromatography-tandem mass spectrometer, or it can provide online analysis services in the form of a cloud platform.

[0019] A system for predicting the risk of developing atopic dermatitis in a subject, according to a specific embodiment of the present invention. Clinical variables of the subjects were obtained, including one or more of the following: parental age, parental history of allergic diseases, maternal income, maternal BMI in early pregnancy, newborn sex, gestational diabetes, and history of gestational hypertension. Parental age refers to the age of both the father and mother, and parental history of allergic diseases refers to the history of both the father and mother's allergic diseases.

[0020] The "preset threshold" serves as a boundary between high and low risk, typically determined using statistical methods (such as the Youden index) based on training set data. The "risk probability value" is a continuous numerical value between 0 and 1; a higher value indicates a higher likelihood of the subject having Alzheimer's disease (AD). The risk assessment module outputs a high-risk warning when the risk probability value is greater than the preset threshold and a low-risk warning when the risk probability value is less than or equal to the preset threshold. For example, in this embodiment, the preset threshold is 0.5; a high-risk warning is output when the risk probability value is greater than 0.5, and a low-risk warning is output when the risk probability value is less than or equal to 0.5.

[0021] This invention can also provide "auxiliary diagnostic conclusions," which can be displayed as text prompts, color-coded indicators, or audible alarms; the specific format can be flexibly configured according to the implementation scenario. The conclusions output by this system are not final diagnoses, but rather supplementary opinions for clinicians to consider in determining whether further detailed and comprehensive examinations are necessary.

[0022] According to a specific embodiment of the system of the present invention, the random forest algorithm is obtained by feature selection from the biomarker content data and clinical variables, and is constructed using the biomarker content data and the clinical baseline variables as feature information; wherein, the model is evaluated by repeated stratified 5-fold cross-validation, and its AUROC is not less than 0.80, and its AUROC verified by an independent external validation cohort is not less than 0.75.

[0023] The "model computation module" is the core algorithm engine of the system. Random forest is an ensemble learning method that constructs a large number of decision trees and fuses the results using majority voting. Each tree is trained on bootstrap samples from the training set, and a subset of features is randomly selected at each split node to enhance model diversity and reduce overfitting.

[0024] The system according to a specific embodiment of the present invention further includes a Luminex liquid phase chip analysis submodule and an ELISA enzyme-linked immunosorbent assay analysis submodule, configured as follows: Umbilical cord blood serum samples were collected from the subjects, and the biomarkers were quantitatively detected to obtain the content data of the biomarkers. If a biomarker was below the detection limit in more than 70% of the samples, it was not included in the subsequent model.

[0025] The "5-fold cross-validation" refers to randomly dividing the training set into five equal parts, using four parts to train the model and one part to validate the model in turn, repeating this process five times, and then taking the average performance index as the evaluation result for that parameter combination. 5-fold cross-validation can effectively prevent overfitting and make full use of limited data samples. The "Area Under the Receiver Operating Characteristic Curve" (AUC-ROC) is a commonly used metric for evaluating the overall performance of binary classification models. An AUC value between 0.5 (no discriminative ability) and 1.0 (perfect discrimination) indicates that the higher the value, the stronger the model's ability to distinguish between positive and negative samples.

[0026] The sample collection, biomarker detection, data preprocessing, model calculation, and result interpretation involved in the above process can all be completed in a standardized and automated manner in medical testing laboratories equipped with the corresponding equipment (Luminex liquid chromatography-microarray analyzers and ELISA enzyme-linked immunosorbent assay analyzers) and computer systems. The entire process does not rely on the operator's subjective experience and has good reproducibility and clinical application value.

[0027] The beneficial effects of this invention are: This invention establishes a method for predicting the risk of atopic dermatitis in infants and young children, which can be completed early after birth, by collecting umbilical cord blood serum samples at birth, screening candidate proteins, quantitatively detecting immune-related biomarkers, variable screening, machine learning modeling, external validation, and model simplification. This method not only has good predictive performance and external generalization ability, but also possesses the feasibility for further simplification and clinical translation.

[0028] First, this invention uses umbilical cord blood serum at birth as the detection target, which can complete the risk assessment of atopic dermatitis in infants and young children before the onset of the disease, thus advancing the time point for identifying high-risk individuals.

[0029] Second, by jointly modeling umbilical cord blood immune-related biomarkers with clinical variables, the predictive performance was significantly improved. Specifically, the AUROC of the AD incidence prediction model increased from 0.560 to 0.869, the F1 score increased from 0.477 to 0.747, the accuracy increased from 0.554 to 0.785, and the Brier score decreased from 0.259 to 0.163. This indicates that the present invention not only improves the discrimination ability but also improves the calibration performance. At the same time, the NRI of the present invention is 0.552 and the IDI is 0.206, indicating that it can effectively improve individual risk reclassification.

[0030] Third, the model constructed in this invention was validated by an independent external cohort. The external validation AUROC of the disease occurrence prediction model still reached 0.786, indicating that it has good generalization ability.

[0031] Fourth, the biomarkers selected in this invention have strong pathway specificity and biological interpretability, and can reduce application complexity while maintaining high performance through model simplification, thus making them more suitable for clinical application.

[0032] Fifth, this invention improves the stability and repeatability of results through standardized detection procedures and variable screening rules, and can reduce application complexity while maintaining high performance through model simplification, thus making it more suitable for clinical application. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram of the proteomics screening process for stable differentially expressed proteins.

[0035] Figure 2 This is the ROC curve of the MS3 model.

[0036] Figure 3 This shows the importance of the top 20 features in the MS3-RF model.

[0037] Figure 4 Displays the calibration curves and Brier scores for each model in MS1 ​​and MS3.

[0038] Figure 5 Displays the calibration curve for MS3-RF.

[0039] Figure 6 This shows the trend of ROC of MS3-RF as the number of features included in the model changes. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0041] This invention collects umbilical vein blood from newborns at birth, separates the serum, and stores it at -80℃. Proteomics is used to screen differentially expressed proteins in independent sub-cohorts to obtain stable differentially expressed proteins, and a candidate biomarker set is constructed based on detection feasibility and pathway representativeness. Luminex liquid chromatography-mass spectrometry (LCS) and ELISA are used to quantitatively detect candidate immune-related proteins. Quality control and variable screening are performed on the detection results; if a certain indicator is below the detection limit in more than 70% of the samples, it is not included in the model. The retained biomarker data and clinical variables are jointly input into a machine learning model for training. The machine learning model includes one or more of random forest, XGBoost, support vector machine, logistic regression, and LASSO. Model performance is evaluated through repeated stratified five-fold cross-validation, and the best model is selected based on indicators such as AUROC, F1, Accuracy, Brier score, DCA, NRI, and IDI. The trained and parameter-locked best model is directly applied to an independent external validation cohort to output the individual AD incidence probability and risk stratification results. The optimal incidence prediction model is a random forest model, which maintains good discriminative ability in the independent external cohort. Furthermore, the optimal model was subjected to feature compression using mRMR to form a simplified risk prediction model suitable for clinical applications.

[0042] The reagent kit and detection methods in this embodiment of the invention are as follows: (1) Human ProcartaPlex Mix & Match 2-plex (Invitrogen, PPX-02-MXU66ED) was used to detect the levels of IL-31 and IL-33 in serum samples; (2)Human ProcartaPlex Mix&Match 26-plex (Invitrogen, PPX-26-MXWCYYA) detects: BDNF, CTACK / CCL27, Eotaxin / CCL11, Eotaxin-2 / CCL24, Eotaxin-3 / CCL26, GM-CSF, I-TAC / CXCL11, IFN-γ, IL-10, IL- 13. IL-17A / CTLA-8, IL-2, IL-21, IL-22, IL-3, IL-4, IL-5, IL-6, IP-10 / CXCL10, MDC, MIG / CXCL9, MIP-1α / CCL3, MIP-3α / CCL20, RANTES / CCL5, TARC / CCL17, TNF-α; (3) Human Multiplex Assay Kit (Cloud-Clone, IS135) detection: PES1, NT5C2, TPP1, TIMP3, Periostin.

[0043] (4) Human TGF-β1 ELISA Kit (Laizhi Biotechnology, LEH822-2S); (5) Human CC motif chemokine 14 (CCL14) ELISA Kit (CUSABIO, CSB-EL004777HU); (6) Human Interleukin 26 (IL-26) ELISA Kit (CUSABIO, CSB-E11716h); (7) Human HLA-B associated transcript 3 (BAT3) ELISA Kit (CUSABIO, CSB-EL002567); (8) ELISA Kit for Collagen Type VI Alpha 3 (COL6a3) (Cloud-Clone, SEC148HU).

[0044] Example 1 In this embodiment, all PAEPAD cohort samples were collected from Daxing Teaching Hospital of Capital Medical University. Umbilical vein serum samples were selected from 128 healthy controls and 110 AD patients (50 with persistent flare-ups and 60 with spontaneous remission). The study was approved by the Ethics Committee of Daxing Teaching Hospital of Capital Medical University (Ethics Approval No.: NO.20190614LLKYLX-3-5), and written informed consent was obtained from all participants.

[0045] Persistent Alzheimer's disease (PAD) was defined as onset within 6 months of age with persistent AD episodes at both 12 and 24 months of age, and with the child's guardian confirming more than 5 AD episodes within one year at the 24-month follow-up. Spontaneous remission Alzheimer's disease (SDR) was defined as onset within 6 months of age with no AD episodes at both 12 and 24 months of age, and with the child's guardian confirming no AD episodes within one year at the 24-month follow-up. All participants were recruited from the PAEPAD cohort without pre-selection or stratification based on gender.

[0046] The inclusion criteria for the PAEPAD cohort were as follows: (1) pregnant women who had registered at Daxing Teaching Hospital of Capital Medical University and planned to give birth at that hospital; (2) women aged ≥18 years; (3) gestational age calculated from the last menstrual period at the time of recruitment was less than 14 weeks + 6 days; (4) residents of Daxing District, Beijing, and planned to continue to reside in Daxing District for at least 2 years postpartum. The total number of participants in the PAEPAD cohort exceeded 2,000.

[0047] Exclusion criteria were as follows: (1) planned induced abortion; (2) mother had a rare comorbidity that, as determined by obstetrics, could lead to miscarriage and congenital defects, including but not limited to malignant tumors, congenital heart disease, and single-gene disorders; (3) recurrent miscarriage; (4) either parent had a mental, psychological, or intellectual disability. Removal criteria were: (1) stillbirth; (2) miscarriage (spontaneous or induced abortion); (3) rare comorbidities that developed after enrollment that could make the participant unsuitable to continue participating in the study, including but not limited to malignant tumors, amniotic fluid embolism, eclampsia, and major birth abnormalities in children. The diagnosis of AD patients was based on the Hanifin-Rajka diagnostic criteria for atopic dermatitis.

[0048] The external validation cohort was from Beijing Children's Hospital Shunyi Women and Children's Hospital, and umbilical vein blood serum samples were collected from 19 healthy controls and 25 AD patients (12 with persistent morbidity and 13 with spontaneous remission). The inclusion criteria were: (1) pregnant women who delivered at Shunyi Women and Children's Hospital; (2) women aged ≥18 years; (3) gestational age calculated from the last menstrual period at the time of recruitment was less than 14 weeks + 6 days; (4) residents of Shunyi District and planning to continue to live in Shunyi District for at least 2 years after delivery. The exclusion criteria were as follows: (1) planned induced abortion; (2) mother with rare comorbidities that, as determined by obstetrics, may lead to miscarriage and congenital defects, including but not limited to malignant tumors, congenital heart disease and single-gene diseases; (3) recurrent miscarriage; (4) either of the prospective parents has a mental, psychological or intellectual disability. Exclusion criteria: (1) stillbirth; (2) miscarriage (spontaneous or induced abortion); (3) rare comorbidities that may make the participant unsuitable to continue participating in the study after enrollment, including but not limited to malignant tumors, amniotic fluid embolism, eclampsia and major birth abnormalities in children.

[0049] In this embodiment, the model establishment cohort was derived from the PAEPAD birth cohort, which included 238 newborns, of whom 128 were healthy controls and 110 were infants with atopic dermatitis (AD). AD was diagnosed according to the Hanifin-Rajka diagnostic criteria for atopic dermatitis, and the target outcome was the occurrence of AD during infancy. The model establishment cohort samples were collected from the Daxing Teaching Hospital of Capital Medical University; the external validation cohort was derived from the Shunyi Women and Children's Hospital of Beijing Children's Hospital, which included 44 newborns.

[0050] Sample collection: When collecting umbilical vein blood from an infant, after removing the placenta, the umbilical cord is clamped with two hemostatic forceps to separate it from the fetus, and the umbilical cord is disinfected with povidone-iodine. A puncture needle is inserted into the umbilical vein, and blood is collected using a blood collection tube without anticoagulants. After centrifugation, the supernatant is extracted, and the sample is then stored at -80°C until analysis.

[0051] Example 2: Proteomics 2.1 Enrichment of low-abundance proteins Low-abundance proteins were enriched using a blood low-abundance protein enrichment kit (Omicsolution, OSFP0002). The specific procedures were as follows: Magnetic microparticle suspension was dispersed evenly by pipetting or sonication. 40 μl of the magnetic microparticle suspension was transferred to a 96-well plate and placed on a magnetic rod for 2 min. The supernatant was discarded after magnetic separation. 100 μl of magnetic microparticle washing buffer was added, and the plate was placed in a thermostatic mixer and vortexed for 5 min (1000 rpm, room temperature). The plate was then placed on a magnetic rod for 2 min, and the supernatant was discarded after magnetic separation. The magnetic beads were resuspended in 100 μL of magnetic microparticle washing buffer, and 100 μL of serum was added. The plate was then incubated in a thermostatic mixer at 37°C with shaking for 1 h (1000 rpm). After incubation, place the 96-well plate on a magnetic bar and let it stand for 2 minutes. Magnetic separation is performed, and the supernatant is discarded. Add 300 μL of magnetic microparticle washing solution and shake to wash for 5 minutes (1000 rpm). Repeat the washing three times. The low-abundance protein enriched on the magnetic beads is the result.

[0052] 2.2 Protein quality testing Prepare BSA standard protein solutions according to the Bradford Protein Quantification Kit instructions, with a gradient concentration range of 0–0.5 g / L. Add BSA standard protein solutions and sample solutions of different dilutions to 96-well plates, making a total volume of 20 μL, with each gradient repeated three times. Quickly add 180 μL of G250 staining solution to each well, let stand at room temperature for 5 minutes, and then measure the absorbance at 595 nm. Plot a standard curve using the absorbance of the standard protein solutions and calculate the protein concentration of the samples. Load 20 μg of protein sample onto a 12% SDS-PAGE gel for electrophoresis. Run the stacking gel at 80 V for 20 min and the separating gel at 120 V for 90 min. Stain the gel with Coomassie Brilliant Blue R-250 and destain until the bands are clearly visible.

[0053] 2.3 Trypsin treatment Take the protein samples obtained in the above experiments, and make up to 100 μL with DB lysis buffer (8 M urea, 100 mM TEAB, pH 8.5). Add trypsin and 100 mM TEAB buffer, mix well, and digest at 37 ℃ for 4 h. Then add trypsin and CaCl2, and continue digestion overnight. Add formic acid to the digestion solution to adjust the pH to below 3, and centrifuge at 12000 g for 5 min at room temperature. Slowly add the supernatant to a C18 desalting column, wash three times with washing buffer (0.1% formic acid, 3% acetonitrile), and then elute with elution buffer (0.1% formic acid, 70% acetonitrile). Collect the eluent of each sample and freeze-dry to obtain peptides.

[0054] 2.4 4D Liquid Crystal Mass Detection The desalted lyophilized peptides were reconstituted in phase A (0.1% formic acid aqueous solution) and then analyzed by LC-MS / MS. The entire system was a timsTOF Pro2 mass spectrometer (Bruker Daltonics) of a Thermo Fisher Scientific, MA, USA, in tandem with an UltiMate 3000 system. A total of 200 ng of sample was loaded (AUR3-15075C18 analytical column: 15 cm × 75 μm id, 1.7 μm particle size, 120 Å pore size, IonOpticks), and the sample was separated by a gradient at 60 min and a column temperature of 50 °C. The column flow rate was controlled at 400 nL / min. The gradient started with 4% B phase (80% acetonitrile, 0.1% formic acid), increasing to 28% within 25 min, 44% within 10 min, reaching 90% within 10 min, maintaining this gradient for 7 min, and then equilibrating at 4% for 8 min. DIA data acquisition was performed using the diaPASEF mass spectrometer in diaPASEF mode, with a scan range of 349-1229 m / z and an isolation window width of 40 Da. During the PASEF MSMS scan, the collision energy increased linearly with ion mobility, from 59 eV (1 / K0 = 1.6 Vs / cm²). 2 The voltage rises to 20 eV (1 / K0 = 0.6 Vs / cm). 2 ).

[0055] 2.5 Identification and Quantification of Proteins DIA data were analyzed using the Spectronaut17 search software with default parameters (BGSFactory Settings (default)) based on the human uniprot protein database. Trypsin digestion was performed, and the search parameters were fixed with Carbamidomethylation (C) 57.02 and variable modification with Oxidation (M) 15.99.

[0056] Protein qualitative standards: Precursor Threshold 1.0% FDR, Protein Threshold 1.0% FDR. The Decoy database was generated using a mutated strategy. Spectronaut performed automatic correction, and local normalization was used for data normalization. Peptides with an FDR less than 1.0% were quantified using MaxLFQ.

[0057] 2.6 Functional Analysis Interproscan software was used for GO and IPR functional annotation (including Pfam.PRINTS.ProDom.SMART, ProSite, and PANTHER databases). COG and KEGG were used to analyze the functional protein families and pathways of the identified proteins. Volcano plot analysis, cluster heatmap analysis, and pathway enrichment analysis by GO, IPR, and KEGG were performed on DEP. STRING DB software was used to predict potential protein-protein interactions.

[0058] In the candidate biomarker screening phase, DIA proteomics was first used to analyze differentially expressed proteins in the PAEPAD cohort and validation cohort. The two independent sub-cohorts identified 44 and 478 differentially expressed proteins, respectively. Cross-matching yielded 7 stable differentially expressed proteins, such as... Figure 1 As shown, this serves as a source of subsequent candidate biomarkers.

[0059] The seven stable differentially expressed proteins mentioned above are: PES1, BAT3 / BAG6, NT5C2, UNC45A, TPP1, COL6A2, and TIMP3.

[0060] Example 3: Quantitative Analysis Experiment of Biomarkers In addition to the seven differentially expressed proteins mentioned above, the inventors further screened multiple cytokines / chemokines and related proteins in serum based on experience, including BDNF, CTACK / CCL27, Eotaxin / CCL11, Eotaxin-2 / CCL24, Eotaxin-3 / CCL26, GM-CSF / CSF2, I-TAC / CXCL11, IFN-γ / IFNG, IL10, IL13, IL17A, IL2, IL21, and IL22. , IL3, IL4, IL5, IL6, IP-10 / CXCL10, MDC / CCL22, MIG / CXCL9, MIP-1α / CCL3, MIP-3α / CCL20, RANTES / CCL5, CCL17 / TARC, TNF-α / TNF, IL31, IL33, IL26, IgE / IGHE, TGF-β1 / TGFB1, CCL14, COL6A3, POSTN / Periostin.

[0061]

[0062]

[0063] Cytokines / chemokines and related proteins in the serum samples were quantified using Luminex liquid-phase chip multifactor detection technology and ELISA. UNC45A and COL6A2 were not included in the model because no corresponding kits were available for their quantification.

[0064] The test samples were serum samples. After collection, the samples were centrifuged according to a standardized procedure, and the supernatant serum was aliquoted and stored at -80℃ until testing.

[0065] 3.1 All Luminex experiments were performed strictly in accordance with the instructions of the respective reagent kit manufacturers and on the same detection platform. The overall workflow included: (1) Preparation of reagents and buffers: Prepare / reconstitute standards, quality control products (if any), detection antibodies / magnetic beads, washing solution, diluent and reading buffer according to the instructions; restore all reagents to the specified temperature and mix well before use.

[0066] (2) Standard curve and control settings: Set a standard curve (multi-point gradient) for each item and set blank wells / background wells; set necessary quality control wells (such as quality control samples provided by the kit or set by the detection platform) according to the batch.

[0067] (3) Sample addition and incubation: Distribute the magnetic bead working solution into the reaction wells / well plates, add the standard, control and test sample, and incubate and shake to mix according to the instructions.

[0068] (4) Washing steps: The magnetic beads are washed using a magnetic separation method to remove unbound components and reduce non-specific background. The number of washes and the washing volume shall be performed in accordance with the instruction manual, and the supernatant shall be thoroughly removed after each wash.

[0069] (5) Detection antibody and colorimetric system reaction: Add detection antibody / label (such as biotinylated antibody, streptavidin-phycoerythrin, etc., subject to the kit instructions) in sequence, incubate under the specified conditions and then perform magnetic washing.

[0070] (6) Reading the data: Resuspend the magnetic beads in reading buffer, mix thoroughly, and then collect the fluorescence signal on a Luminex 200. The data acquisition parameters (number of beads per well, reading volume, PMT settings, etc.) are set uniformly by the detection platform according to the kit requirements.

[0071] (7) Concentration calculation: Data processing was performed using ProcartaPlex Analyst 1.0 software, including standard curve fitting, sample concentration conversion, dilution factor calculation, and output result table. The standard curve fitting model was executed according to the software / reagent default settings (e.g., 5PL / 4PL, depending on the platform settings), and the curve fitting quality was checked.

[0072] 3.2 The ELISA experiment was performed according to the instructions for each kit. The overall steps included: (1) Reagent preparation: reconstitution and serial dilution of standards, preparation of working solutions (washing solution, diluent, enzyme conjugate, substrate solution, stop solution, etc.), and equilibration of all reagents to the specified temperature as required.

[0073] (2) Adding samples and reaction: Add standards, blank control and samples (repeat wells can be set) to the pre-coated enzyme plate, and incubate under the specified conditions to allow the target molecules to bind to the coated antibody.

[0074] (3) Washing: Wash the plate according to the number of washes and the volume specified in the instruction manual to reduce non-specific background; ensure that each wash fully removes residual liquid.

[0075] (4) Add detection antibody / enzyme conjugate: Add detection antibody or enzyme conjugate according to the kit procedure, incubate and wash the plate again.

[0076] (5) Color development and termination: Add substrate solution to develop color, and terminate the reaction within the specified time.

[0077] (6) Reading: Use BioTek 800 TS to read the absorbance at the specified wavelength (the specific wavelength is specified by the kit), and use Gen5 software to fit the standard curve and convert the concentration.

[0078] (7) Concentration back-calculation: Back-calculate based on the sample dilution factor and output the final concentration result.

[0079] Standard curve: A standard curve gradient point was set for each experiment to cover the detection range specified by the kit; curve fitting was performed using Gen5 software, and the fitting model was selected and recorded according to the kit / software recommendations.

[0080] Repeat wells: Set up repeat wells where feasible to assess intra-batch repeatability; if there are significant differences in repeat wells, check according to the platform's quality control process and decide whether to retest.

[0081] Background and blank control: Set blank wells / negative control wells to monitor background signals; if the background rises abnormally, check the plate washing, reagent preparation and operation records.

[0082] Batch consistency: Complete the testing of key samples under the same reagent lot number or the same testing conditions as far as possible; if cross-batch testing is required, record the batch information for subsequent analysis or sensitivity inspection.

[0083] Variable availability screening is performed after both Luminex and ELISA tests: if the detected value of a certain index to be tested is lower than the lower limit of quantitation (lower than LLOQ / LOD, specifically as defined by the kit) in more than 70% of the samples, this index has insufficient information in this study and will not be included in the model as an independent variable.

[0084] For indexes that do not meet the above exclusion criteria but have partial samples with values below the lower limit of quantitation, the data below the lower limit are output by the testing platform according to unified rules (for example, marked as "<LLOQ" or given as a software-calculated value), and are processed and retained in subsequent statistical processing according to pre-set data cleaning rules to ensure the consistency and traceability of data processing.

[0085] After calculation and screening, the present invention screens out 23 biomarker variables from the initial 39 cytokines / chemokines and related proteins to be included in the model. The expression levels of detectable biomarkers in the cord blood of healthy control infants and infants with atopic dermatitis are shown in the following table: Table 1 Expression levels of biomarkers in cord blood of healthy control infants and infants with atopic dermatitis

[0086] Example 4 Machine Learning Model In this example, the following machine learning models are constructed, and five classification algorithms are adopted in total: Random Forest (RF, n_estimators = 300), eXtreme Gradient Boosting (XGBoost, n_estimators = 300, learning rate = 0.05), Support Vector Machine (SVM, linear kernel), Logistic Regression (LR), and L1-penalized Logistic Regression (LASSO). All analyses are completed in the Python 3.9 environment, implemented using the scikit-learn and xgboost software packages. All analyses are implemented in the Python environment (using the scikit-learn, xgboost, imbalanced-learn and pymrmr libraries).

[0087] 4.1 Random Forest (RF) The RF model uses 300 decision trees, with Gini impurity as the splitting criterion. The maximum tree depth, minimum number of split samples, and feature subset size are determined through grid search and five-fold cross-validation. Variable importance is quantified using the mean decrease in Gini.

[0088] 4.2 Extreme Gradient Boosting (XGBoost) The XGBoost model contains 300 trees, with a learning rate of 0.05 and a maximum tree depth of 4. The subsample ratio and feature column sampling ratio are optimized using cross-validation, and early stopping is used to determine the optimal number of iterations. Feature importance is evaluated using gain and SHAP (Shapley Additive Explanations) values.

[0089] 4.3 Support Vector Machine (SVM) An SVM classifier with a linear kernel was used, as preliminary tests showed that non-linear kernels did not significantly improve performance. The regularization parameter C was optimized through grid search to balance maximizing the margin with classification error. Z-score normalization was performed on the data before model training to ensure consistency across feature scales.

[0090] 4.4 Logistic Regression (LR) Logistic regression serves as the baseline linear classifier, describing the linear relationship between independent and outcome variables using a log-odds function. To prevent overfitting and stabilize parameter estimates in the presence of collinear features, regularized logistic regression with an L2 penalty term (in ridge regression form) is employed. The regularization strength parameter is adjusted using five-fold cross-validation. The standardized model coefficients can be interpreted as the log-odds ratio of each predictor.

[0091] 4.5 LASSO Logistic Regression LASSO (Least Absolute Shrinkage and Selection Operator) incorporates an L1 penalty term into the objective function to compress the coefficients, achieving simultaneous regularization and feature selection. This study is based on a logistic regression model with an L1 penalty term, using the liblinear solver. The optimal regularization parameter λ is determined through internal cross-validation to minimize the mean log-loss. In the final model, features with non-zero coefficients are considered variables with predictive significance.

[0092] Data was stratified and sampled at a ratio of 80% training set to 20% test set. The LR, SVM, and LASSO models were standardized. MS1 was pre-defined as a model set containing only clinical variables (parental age, parental history of allergies, maternal income, maternal early pregnancy BMI, newborn sex, history of gestational diabetes, and gestational hypertension), with candidate predictors limited to clinical variables. The set of clinical variables remained constant across all cross-validation tradeoffs, and no data-driven feature selection was applied. To improve model performance, 23 selected biomarkers were introduced to construct MS3, which is a model set containing both clinical variables and biomarkers.

[0093] Model performance was evaluated using repeated stratified five-fold cross-validation (repeated 10 times) on the training set. Metrics included Area Under the Receiver Operating Characteristic curve (AUROC), accuracy, and F1 score. Calibration curves were generated using a bootstrap method to generate 95% confidence intervals, and Brier scores, calibration slopes, and intercepts were reported. Clinical usability was further evaluated using DCA, and the number of intervention reductions per 100 patients was reported. Models were compared using the continuous net reclassification index (NRI) and the integrated discriminant improvement index (IDI). To assess model robustness, repeated training analyses with different random seeds were performed. A fixed random seed was used, and version control was employed to ensure reproducibility. The classifier with the highest AUROC median was selected as the optimal model: RF for disease occurrence prediction and LASSO for disease persistence prediction. To obtain more concise models, we performed minimumRedundancy Maximum Relevance (mRMR) feature ranking on these classifiers on the discovery set, generating M5 and M6 models respectively. Feature selection used the minimum redundancy maximum relevance mRMR algorithm, and the model was retrained after selecting the top 20 predictive features. All preprocessing steps (normalization parameters, batch correction parameters, feature selection, and decision thresholds determined based on Youden's J from the discovery set) were frozen before testing.

[0094] In external validation, the best-performing locked model, MS3-RF, was directly applied to an independent external cohort without refitting. AUROC, accuracy, and F1 score are reported.

[0095] For predicting AD incidence, models that only include clinical variables are named MS1, and models that include biomarkers are named MS3. The AUROC for MS3 is [see...]. Figure 2 The classifier with the highest AUROC median is selected as the optimal model. In disease occurrence prediction, Random Forest (MS3-RF) in MS3 is the optimal classifier. The importance of its top 20 features is shown in [link to relevant documentation]. Figure 3 The x-axis represents the importance of the feature, and the y-axis represents the name of the feature. The larger the value, the greater the relative contribution of the feature to the model's prediction (regardless of whether it is positive or negative).

[0096] The calibration curves and Brier scores for each model in MS1 ​​and MS3 are shown below. Figure 4 The horizontal axis represents the average predicted probability after grouping, and the vertical axis represents the proportion of actual occurrence of atopic dermatitis in the corresponding group. The black dashed line represents the ideal calibration state (the predicted probability is completely consistent with the actual probability of occurrence). The Brier score corresponding to each model is marked in the legend, which is used to quantify the overall calibration error of the model; the smaller the value, the better the calibration performance.

[0097] The results of the decision curve analysis are shown below. Figure 5 The horizontal axis represents the predicted probability of atopic dermatitis, and the vertical axis represents the actual proportion of occurrence within the corresponding probability interval. The dashed line represents the ideal calibration state (the predicted probability is completely consistent with the actual probability), and the solid line is the MS3-RF calibration curve, used to evaluate the consistency between the model's predicted probability and the actual outcome.

[0098] The results showed that the AUROC of the MS3-RF model was 0.869, which was significantly higher than that of the MS1 baseline model containing only clinical variables (AUROC 0.560); the F1 score of the MS3-RF model was 0.747, which was higher than that of the baseline model (0.477); the accuracy was 0.785, which was higher than that of the baseline model (0.554); and the Brier score of the MS3-RF model was 0.163, which was better than that of the baseline model (0.259).

[0099] Compared to the baseline model, the MS3-RF had a continuous net reclassification index (NRI) of 0.552 and an integrated discrimination improvement index (IDI) of 0.206, indicating that by introducing umbilical cord blood immune-related biomarkers, this invention can significantly improve the discrimination ability, calibration performance, and risk reclassification ability of predicting the incidence risk of AD in infants and young children.

[0100] During the external validation phase, the model was not refitted, the variables were not re-selected, and the threshold (0.5) was not reset. Instead, the MS3-RF model locked in the development queue was directly applied to the independent external validation queue (Beijing Children's Hospital Shunyi Women and Children's Hospital, 44 cases).

[0101] The results show that MS3-RF has an AUROC of 0.786 in the external validation queue, maintaining good discriminative ability and good classification performance, indicating that the disease occurrence prediction model established in this invention has good generalization ability.

[0102] To improve the ease of clinical implementation, the minimum redundancy maximum relevance (mRMR) method is further used for feature compression based on the complete model.

[0103] The trend of AUROC of the mRMR model with the number of included features is shown in the figure. Figure 6 The x-axis represents the number of features included in the model, and the y-axis represents the ROC-AUC. The solid line represents the average ROC-AUC of the model under different numbers of features, and the shaded area represents the uncertainty range (95% CI). The results show that after model simplification, the disease occurrence prediction model still maintains good performance, with an AUROC of 0.846. This indicates that even with a reduction in the number of features, the present invention can still maintain high predictive efficacy, which is beneficial for subsequent clinical application.

[0104] In summary, this embodiment establishes a method for predicting the risk of atopic dermatitis in infants and young children that can be completed early after birth. This method involves collecting umbilical cord blood serum samples at birth, screening candidate proteins, quantitative detection of immune-related biomarkers, variable screening, machine learning modeling, external validation, and model simplification. The method not only has good predictive performance and external generalization ability, but also possesses the feasibility for further simplification and clinical translation.

[0105] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. The application of a biomarker in the preparation of products for predicting the risk of developing atopic dermatitis in infants and young children, characterized in that, The biomarkers are one or more combinations of BDNF, CTACK / CCL27, Eotaxin / CCL11, Eotaxin-2 / CCL24, Eotaxin-3 / CCL26, GM-CSF / CSF2, I-TAC / CXCL11, IL-2, IL-10, IL-21, IP-10, MDC / CCL22, MIG / CXCL9, MIP-1α / CCL3, MIP-3α / CCL20, RANTES / CCL5, BAT3 / BAG6, TGF-β1 / TGFB1, CCL14, COL6A3, TIMP3, POSTN / Periostin, and CCL17 / TARC.

2. The application according to claim 1, characterized in that, The product is a reagent kit, chip, reagent, or monitoring system for detecting biomarkers.

3. A kit for predicting the risk of developing atopic dermatitis in infants and young children, characterized in that, The kit includes reagents for detecting biomarker content data, wherein the biomarkers include one or more combinations of BDNF, CTACK / CCL27, Eotaxin / CCL11, Eotaxin-2 / CCL24, Eotaxin-3 / CCL26, GM-CSF / CSF2, I-TAC / CXCL11, IL-2, IL-10, IL-21, IP-10, MDC / CCL22, MIG / CXCL9, MIP-1α / CCL3, MIP-3α / CCL20, RANTES / CCL5, BAT3 / BAG6, TGF-β1 / TGFB1, CCL14, COL6A3, TIMP3, POSTN / Periostin, and CCL17 / TARC.

4. A system for predicting the risk of developing atopic dermatitis in infants and young children, characterized in that, The system includes a detection module, a model calculation module, and a risk assessment module, wherein... The detection module acquires the following biomarker content data of the subject: BDNF, CTACK / CCL27, Eotaxin / CCL11, Eotaxin-2 / CCL24, Eotaxin-3 / CCL26, GM-CSF / CSF2, I-TAC / CXCL11, IL-2, IL-10, IL-21, IP-10, MDC / CCL22, MIG / CXCL9, MIP-1α / CCL3, MIP-3α / CCL20, RANTES / CCL5, BAT3 / BAG6, TGF-β1 / TGFB1, CCL14, COL6A3, TIMP3, POSTN / Periostin, CCL17 / TARC; The model calculation module, based on the obtained biomarker content data and clinical variables, performs calculations using a random forest model to obtain the risk probability value of the subject's atopic dermatitis. The risk determination module is used to compare the risk probability value with a preset threshold and generate an auxiliary diagnostic conclusion based on the comparison result.

5. The system according to claim 4, characterized in that, The detection module also includes a Luminex liquid phase chip analysis submodule and an ELISA enzyme-linked immunosorbent assay submodule to obtain biomarker content data.

6. The system according to claim 4, characterized in that, The risk determination module is used to output a high-risk warning when the risk probability value is greater than the preset threshold, and to output a low-risk warning when the risk probability value is less than or equal to the preset threshold.

7. The system according to claim 4, characterized in that, The data processing module also includes a model performance evaluation unit, which uses repeated stratified five-fold cross-validation to evaluate model performance. The evaluation metrics include the area under the receiver operating characteristic curve, accuracy, and F1 score.

8. The system according to claim 4, characterized in that, Umbilical cord blood serum samples were obtained from the subjects, and the biomarkers were quantitatively detected.

9. The system according to claim 4, characterized in that, The clinical variables included parental age, parental history of allergic diseases, maternal income, maternal BMI in early pregnancy, newborn sex, gestational diabetes, and history of gestational hypertension.

Images