Immune metabolism crosstalk multi-omics anatomical Alzheimer disease prediction method and device

Abstract

The invention discloses a multi-omics anatomical Alzheimer's disease prediction method and device based on immune metabolic crosstalk. The invention relates to application of a reagent for detecting a gene combination in preparation of a product for diagnosing Alzheimer's disease. The gene combination comprises Uqcrb, Klf6, Txn, Lsm5, Psmc1, Eif3e and Laptm5. The invention further discloses a kit. The prediction method comprises the following steps: acquiring transcriptome data and single cell data of an AD patient and a healthy person; performing pretreatment; performing differential gene analysis to obtain a gene analysis result; analyzing expression differences of different metabolic pathways in various cell populations; screening out a gene related to the AD disease; extracting high-expression genes in the first transcriptome data as a feature set; training a plurality of machine learning models by using the training data to obtain three final training models; constructing a dynamic column graph containing the key genes; and predicting AD risk of a single patient.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary application field of bioinformatics, genomics, and machine learning in biomedicine, specifically involving a multi-omics anatomical method and device for predicting Alzheimer's disease by immune metabolic crosstalk. Background Technology

[0002] Alzheimer's disease (AD) is the most common neurodegenerative disease among the elderly, with a complex pathogenesis and a lack of effective early diagnostic methods. Currently, AD diagnosis mainly relies on clinical symptoms, cerebrospinal fluid biomarkers (such as Aβ and Tau proteins), and neuroimaging examinations (such as PET-CT). However, these methods have drawbacks such as high invasiveness, high cost, and limited availability. Furthermore, the pathological mechanisms of AD are not fully understood; hypotheses regarding genetic factors, immune metabolic abnormalities, and pathogen infection require further verification.

[0003] Existing research indicates that the immune system plays a crucial role in the pathogenesis of Alzheimer's disease (AD), but traditional studies often focus on single immune cells or metabolic pathways, lacking systematic analysis. Furthermore, the development of blood-based biomarkers is limited by sample heterogeneity and short-term fluctuations, making stable predictions difficult. Therefore, there is an urgent need for a non-invasive, highly specific predictive model based on multi-omics data to address the clinical challenges in early AD diagnosis. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a comprehensive multi-omics anatomical method and device for predicting Alzheimer's disease based on an immune metabolic cascade. This method and device integrate characteristic gene expression profiles of immune cell metabolic states with machine learning algorithms to achieve early diagnosis and risk prediction of AD, offering advantages such as high specificity, low cost, and clinical applicability.

[0005] In a first aspect, a reagent for detecting a gene combination is provided for use in the preparation of products for diagnosing Alzheimer's disease, wherein the gene combination includes: Uqcrb, Klf6, Txn, Lsm5, Psmc1, Eif3e, and Laptm5.

[0006] In a second aspect, a kit is provided, the kit comprising detection reagents for detecting a combination of genes including: Uqcrb, Klf6, Txn, Lsm5, Psmc1, Eif3e, and Laptm5.

[0007] Thirdly, a comprehensive multi-omics approach to predicting Alzheimer's disease, based on an immune metabolic cascade, is provided, including the following steps: Acquire transcriptomic and single-cell data from AD patients and healthy individuals; The transcriptome data and single-cell data were preprocessed separately to obtain the first transcriptome data and the first single-cell data. Clustering, dimensionality reduction, and pathway enrichment analysis were performed on the first single-cell data to identify nucleic acid metabolism pathways in glial cells and monocytes, as well as amino acid metabolism pathways in neurons and T / B cells; differential gene analysis was conducted to obtain gene analysis results; Based on the positive results obtained from gene analysis, we analyzed the expression differences of different metabolic pathways in various cell populations and screened out genes related to Alzheimer's disease. The first transcribed data was used as training data and divided into a training set and a validation set. Based on AD-sensitive metabolic pathways, highly expressed genes from the first transcriptome data were extracted as a feature set. The AD-related genes of macrophages, protective T cells, and protective B cells related to nucleotide metabolism in the feature set were used as factors; multiple machine learning models were trained using the training data to obtain three final training models; Evaluate the performance of multiple trained models and select the optimal model; Based on the optimal model, a dynamic nomogram containing key genes is constructed to predict the risk of AD in individual patients. Fourthly, it provides a comprehensive multi-omics anatomical Alzheimer's disease prediction device based on the immune metabolic chain, including: The acquisition module is used to acquire transcriptomic and single-cell data from AD patients and healthy individuals. The preprocessing module is used to preprocess the transcriptome data and single-cell data respectively to obtain the first transcriptome data and the first single-cell data; The analysis module is used to perform clustering, dimensionality reduction, and pathway enrichment analysis on the first single-cell data to identify nucleic acid metabolism pathways in glial cells and monocytes, as well as amino acid metabolism pathways in neurons and T / B cells; differential gene analysis is used to obtain gene analysis results; The gene screening module is used to analyze the expression differences of different metabolic pathways in various cell populations based on the positive results obtained from gene analysis; and to screen out genes related to Alzheimer's disease. The partitioning module is used to divide the first transcribed data into a training set and a validation set, using the training data as training data. The extraction module is used to extract highly expressed genes from the first transcriptome data as a feature set based on AD-sensitive metabolic pathways. The training module uses AD-related genes from macrophages, protective T cells, and protective B cells in the feature set as factors; it uses the training data to train various machine learning models, resulting in three final training models. The model selection module is used to evaluate the performance of multiple trained models and select the optimal model. The prediction module is used to construct dynamic nomograms containing key genes based on the optimal model; and to predict the risk of AD in individual patients.

[0008] Fifthly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements a comprehensive multi-omics anatomical Alzheimer's disease prediction method based on the immune metabolic chain as described in the third aspect.

[0009] In a sixth aspect, a non-transitory computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the integrated multi-omics anatomical Alzheimer's disease prediction method based on the immune metabolic chain as described in the third aspect.

[0010] In a seventh aspect, a computer program product is provided, comprising a computer program that, when executed by a processor, implements a comprehensive multi-omics anatomical Alzheimer's disease prediction method based on an immune metabolic chain as described in the third aspect.

[0011] The beneficial effects of this invention are as follows: The method and apparatus described in this invention combine immune metabolic status with multi-omics data to reveal the molecular mechanisms of Alzheimer's disease (AD). They also enable non-invasive detection based on blood transcriptomics and utilize dynamic nomogram tools for real-time risk assessment. This method and apparatus are inexpensive and widely applicable, particularly in primary healthcare settings, demonstrating promising application prospects and significant potential for widespread adoption. Attached Figure Description

[0012] Figure 1 To construct and validate a machine learning-based Alzheimer's disease (AD) prediction model; among which, Figure 1 (a) To determine the optimal set of differentially expressed genes by intersection analysis of metabolic state-related genes (high metabolic activity group vs. low metabolic activity group in blood single cells) and blood transcriptome differences (AD vs. normal control [NC]). Three characteristic gene sets were obtained: risk monocytes (Risk-M, high NamIis monocytes), protective T cells (Protective-T, high Aam T cells), and protective B cells (Protective-B, high Aam B cells).

[0013] Figure 1 (b) Cross-dataset partitioning strategy. Transcriptome microarray data from GSE63060, GSE63061 and the Chinese cohort were randomly divided into 10 parts, of which 7 / 10 were used for model training and 3 / 10 were used for independent validation.

[0014] Figure 1 (c) is a heatmap of the area under the receiver operating characteristic (AUC) curve of the top model constructed based on the Risk-M, Protective-T and Protective-B gene sets.

[0015] Figure 1 (d) shows the performance comparison, where the Elastic Network Regression [α=0.6] + Random Forest (Enet[alpha=0.6] + RF) hybrid model achieves the best predictive performance when using the Protective-B gene set.

[0016] Figure 1 (e) ROC analysis shows that the model has consistent predictive performance across different datasets: training set (AUC=1.000, 95% CI: 0.999–1.000), Chinese cohort (AUC=0.949), GSE63060 (AUC=0.828), and GSE63061 (AUC=0.841).

[0017] Figure 1 (f) shows the decision curve analysis, indicating that when the threshold probability is >0.1, the model has a significant net clinical benefit (red line) compared to the "all treatment" or "no treatment" strategy (black line).

[0018] Figure 1 (g) shows that the predicted probability is highly consistent with the observed result (the deviation correction curve almost coincides with the ideal line).

[0019] Figure 1 (h) shows the expression patterns of seven core model genes (Laptm5, Klf6, Eif3e, Psmc1, Txn, Lsm5, and Uqcrb) in AD blood samples, all of which showed significant differences compared to NC. P<0.01, P<0.001).

[0020] Figure 1 (i) The SHAP summary plot shows the importance of features and the direction of effects. Laptm5 (SHAP value range: -0.15~0.2) and Klf6 (-0.2~0.2) are key bidirectional regulators.

[0021] Figure 1 (j) shows the predictive drivers in the SHAP action plot (representative sample): Laptm5 (value = 11.4) and Eif3e (11.1) contribute positively (red arrows), while Klf6 (9.9) and Psmc1 (9.84) have negative effects (blue arrows), with a final output probability of 0.102 (baseline = 0.556).

[0022] Figure 2 To validate the 5xFAD mouse model (including the pre-symptom stage); among which, Figure 2 (a, j) is a schematic diagram of the experimental design for 9-month-old (9 m) and 2.5-month-old (2.5 m) 5xFAD mice and age-matched wild-type (WT) controls.

[0023] Figure 2 (b, k) represents the escape latency (time required to locate the exit) for two genotype mice in the Morris water maze test.

[0024] Figure 2 (c, l) represents the number of times the platform area is traversed in the water maze.

[0025] Figure 2 (d, m) represents the dwell time in the target platform area (memory exploration test).

[0026] Figure 2 (e, n) represents the average swimming speed of the mice during the test.

[0027] Figure 2 (f, o) represents the behavioral trajectories of WT and 5xFAD mice on the final test day (*P<0.05, **P<0.01 vs WT; ns had no statistical significance).

[0028] Figure 2 Brain slices stained with 6E10 (green), Aβ (red), and HST (blue) (g, p) show amyloid plaques in 5xFAD mice.

[0029] Figure 2 (h) shows the ROC curve (AUC=1.0) of the logistic regression model based on 7 differentially expressed genes (Klf6, Uqcrb, Eif3e, Txn1, Lsm5, Psmc1, Laptm5) for 9 m-5xFAD mice, indicating excellent diagnostic efficacy in the symptomatic phase.

[0030] Figure 2 (i) Single-gene ROC analysis of 9 m-5xFAD mice: Lsm5 had the highest predictive value (AUC=0.815).

[0031] Figure 2 (q) is the ROC validation of the model in pre-symptom 2.5 m-5xFAD mice (AUC=0.922), demonstrating its prospective predictive value for familial AD. Sample size: 9 m cohort, WT n=6, 5xFAD n=9; 2.5 m cohort, WT n=8, 5xFAD n=8.

[0032] Figure 3 This reflects the disease specificity and metabolic robustness of the AD prediction model; among which, Figure 3 (a) AD cohort: The Protective-B model showed excellent diagnostic performance (AUC=0.805, 95% CI: 0.763-0.846). Single biomarker ROC analysis confirmed that all 7 core genes were diagnostically relevant (e.g., Klf6 AUC=0.716; Uqcrb AUC=0.688).

[0033] Figure 3 (b) Parkinson's disease (PD) cohort: model performance was significantly reduced (AUC=0.637, 95% CI: 0.591-0.686), and no single biomarker showed diagnostic value (all AUC<0.613), confirming specificity for PD.

[0034] Figure 3 (c) Frontotemporal dementia (FTD) cohort: limited diagnostic utility (AUC=0.603, 95% CI: 0.560–0.644). Biomarkers showed no discriminatory power (Eif3e had the highest AUC=0.580), ruling out the possibility of misclassification of FTD.

[0035] Figure 3 (d) Obese cohort (BMI>30): The model maintained moderate performance (AUC=0.644, 95% CI: 0.580-0.705), but the predictive value of single biomarkers was negligible (Klf6 had the highest AUC=0.585), indicating its resistance to metabolic confounding.

[0036] Figure 4 This section details the clinical translation and online tool development of AD prediction models; among which, Figure 4 (a) A clinical nomogram integrating the expression of Klf6, Uqcrb, Txn, Eif3e, Lsm5, Psmc1 and Laptm5, used to calculate AD risk scores (e.g., 180 points = 90% probability).

[0037] Figure 4 (b) The logistic regression results of the 7-gene model showed that Klf6 was a strong protective factor (OR=37.7, P<1.91e-08) and Laptm5 was a risk factor (OR=0.047, P<2.91e-06).

[0038] Figure 4 (c) The simplified 4-gene model showed better parsimony (AIC 453.94 vs 455.56), and Klf6 remained the strongest protective factor (OR=60.4, P<1.11e-11).

[0039] Figure 4 (d) Klf6, Txn, Laptm5 and Psmc1 were retained in the final 4-gene model, among which Txn showed a protective effect (OR=0.35, P<3.09e-06).

[0040] Figure 4 (e) ROC analysis of the 4-gene model in independent validation (AUC=0.802, 95% CI: 0.757-0.844).

[0041] Figure 4 (f) The diagnostic performance of single genes ranged from Klf6 (AUC=0.716) to Laptm5 (AUC=0.575).

[0042] Figure 4 (g) indicates that the decision curve analysis shows clinical efficacy when the threshold probability is >10%.

[0043] Figure 4 (h) shows excellent prediction accuracy as a calibration curve (Hosmer-Lemeshow P=0.62).

[0044] Figure 4 (ij) is the interface of a network-based real-time AD risk prediction tool, showing gene expression input examples for NC(i) and AD(j) populations.

[0045] Figure 5 A schematic diagram of the structure of the integrated multi-omics anatomical Alzheimer's disease prediction device with immune metabolic crosstalk provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0046] The present invention will be further described below with reference to specific embodiments, but the content of the present invention is not limited thereto.

[0047] Existing research on early diagnosis and prediction of Alzheimer's disease (AD) focuses on single immune cells or metabolic pathways, lacking systematic analysis. Furthermore, the development of blood-based biomarkers is limited by sample heterogeneity and short-term variability, making stable predictions difficult.

[0048] Therefore, how to integrate multiple metabolic pathways to improve the stability and accuracy of AD prediction is an urgent problem to be solved.

[0049] See Figure 1 (a) is a flowchart illustrating a comprehensive multi-omics approach to predicting Alzheimer's disease based on immunometabolic crosstalk, including the following steps: S100 acquires transcriptomic and single-cell data from AD patients and healthy individuals.

[0050] In one possible implementation, the transcriptome data includes blood single-cell transcriptome data. In another possible implementation, the single-cell data includes brain tissue and peripheral blood single-cell data.

[0051] S200, the transcriptome data and single-cell data are preprocessed separately to obtain the first transcriptome data and the first single-cell data.

[0052] Specifically, transcriptome data from various AD patients were obtained from the Gene Expression Omnibus (GEO) database (URL: https: / / www.ncbi.nlm.nih.gov / geo / info / datasets.html). These datasets include 1) GSE63060: peripheral blood samples from European AD patients, including 104 healthy controls and 145 AD patients; 2) GSE63061: peripheral blood samples from European AD patients, including 135 healthy controls and 139 AD patients.

[0053] To improve sample diversity and localized prediction performance, a Chinese cohort was added.

[0054] Chinese cohort construction and sample processing: Subject inclusion: AD group: 44 patients diagnosed by PET (meeting NINCDS-ADRDA criteria), from Wuhan University People's Hospital, Zhongnan Hospital, and Henan Provincial People's Hospital. Healthy control group: 30 healthy individuals (excluding AD and other dementias), recruited from the community and those who visited the same hospital. All subjects signed informed consent forms. Ethics approval numbers: WHU-LFMD-IRB2024011, HNEECKY-2024-217.

[0055] Sample collection and sequencing: Peripheral blood was collected after a 12-hour fast, and peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll density gradient centrifugation (Solepro reagent). Peripheral blood transcriptome sequencing (RNA-seq) was then performed by the Beijing Genomics Institute (BGI).

[0056] Preprocessing and standardization were performed on the GSE63060, GSE63061, and Chinese cohort data respectively: The limma and ggpubr packages were loaded, and the `read.table` function was used to read the expression matrix file, converting the data to matrix format, setting row names to gene names, removing gene names from the first column, retaining only expression data, and using the `avereps` function to average duplicate values. Quantiles of expression values ​​were calculated to determine if log2 transformation was needed. The `normalizeBetweenArrays` function was used to standardize the arrays. The `read.table` function was used to read the sample information files of the control and experimental groups respectively, extracting the expression data from the healthy control group and the experimental group, and merging them into a single data matrix to obtain the first transcriptome data.

[0057] In one possible implementation, the preprocessing method for the single-cell data includes: reading single-cell expression profile data, filtering low-expression genes, and standardizing and normalizing.

[0058] Specifically, locate the GSE181279 dataset in the GEO database (download link: https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi?acc=GSE181279) and download the Series MatrixFile. After downloading and unzipping, you will get a file containing sample clinical information and gene expression profiles. Download the barcodes, features, and matrix data. Use R packages such as Seurat and Harmony to import the downloaded matrix data into the R software system. Read the single-cell expression profile data, filter low-expression genes, and standardize the data to ensure comparability between different samples. Then, normalize the data to eliminate technical bias.

[0059] S300 performs clustering, dimensionality reduction, and pathway enrichment analysis on the first single-cell data to identify nucleic acid metabolism pathways in glial cells and monocytes, as well as amino acid metabolism pathways in neurons and T / B cells; differential gene analysis is then performed to obtain gene analysis results.

[0060] In one possible implementation, S300 includes: S310. Perform dimensionality reduction and cluster analysis on the first cell data to obtain cluster analysis results including the main feature vectors, the optimal number of principal components, and the uniform manifold approximation and projection analysis.

[0061] Principal Component Analysis (PCA) analysis: Principal component analysis (PCA) is performed on the first cell data to extract the main feature vectors.

[0062] Determine the optimal number of principal components: Use the tool ElbowPlot to observe the optimal number of principal components (pcs).

[0063] Uniform Manifold Approximation and Projection (UMAP) Analysis: Based on the PCA results, UMAP analysis is performed to visualize the clustering results and observe the positions between each cluster.

[0064] S320. Based on the cluster analysis results, perform cell clustering and annotation to identify cell populations that have an important impact on AD.

[0065] Cluster analysis: Based on the UMAP analysis results, the cells were clustered.

[0066] Cell annotation: Clusters were annotated using the Seurat package to identify cell populations that have a significant impact on AD development.

[0067] S330. Cell population regrouping and gene expression level analysis: Cell populations were regrouped; gene expression levels of metabolite pathways in different cell populations were analyzed.

[0068] S340. Extract T cells, B cells, and monocytes from the regrouped cell populations; based on the gene expression levels of metabolite pathways, divide the cells into a high-metabolic immune cell group and a low-metabolic immune cell group.

[0069] It should be noted that the T cells include: T cells obtained from amino acid metabolism enriched by protective metabolite pathways and T cells obtained from amino acid metabolism enriched by dangerous metabolite pathways; the B cells include: B cells obtained from amino acid metabolism enriched by protective metabolite pathways and B cells obtained from amino acid metabolism enriched by dangerous metabolite pathways; the monocytes include monocytes obtained from amino acid metabolism enriched by protective metabolite pathways and monocytes obtained from amino acid metabolism enriched by dangerous metabolite pathways.

[0070] S350, differential gene analysis, to obtain gene analysis results.

[0071] Comparing differentially expressed genes between high and low metabolites: Using Seurat's FindMarkers function, differentially expressed genes were compared between the high and low metabolites.

[0072] Visualizing differentially expressed genes: The expression of differentially expressed genes is shown through volcano plots and heatmaps.

[0073] Based on the positive results obtained from S400 and S350 gene analysis, we analyzed the expression differences of different metabolic pathways in various cell populations; and screened out genes related to AD disease to serve as input factors for machine learning models. Understandably, if no significant differences are found in a particular expression differential analysis, the results will not be used. The final set of differentially expressed genes will serve as a key input factor for subsequent machine learning model construction, used for feature selection and prediction model building.

[0074] In one possible implementation, S400 includes: The differential gene lists of high and low metabolites for each cell type (T / B / monocytes) were analyzed for differential expression using statistical methods; screening criteria were determined to identify genes associated with Alzheimer's disease.

[0075] Specifically, the screening criteria were as follows: significance threshold: adjusted p-value (adj.P.Val) < 0.05; expression change threshold: average log2-fold change > 0.5.

[0076] Finally, differentially expressed genes in monocytes that are associated with dangerous nucleotide metabolism and immune response were identified through screening: RNASET2, ADAR, AIF1, HLA-E, SERPINA1, FGR, PAPSS2, FCER1G, CSF1R, GPX1, and PTPN6.

[0077] Protective genes for B cells: EEF1A1, EEF1B2, EEF1G, CD74, TPT1, FOS, LAPTM5, DDX5, CD69, TOMM7, IL4R, TSC22D3, MTSS1, MEF2C, NME1-NME2, STK17B, ADK, TCL1A, UQCRB, BACH2, MYADM, CD48, CD83, EIF3E, ACTG1, FBL, FCER2, LBH, TXN, JUNB, PPA1, PTPRC , BCLAF1, ZFP36, AIM2, SMS, PSME2, ​​TUBA1A, ZFP36L1, SLC11A1, C1QBP, KLF6, BRD2, FGR, YPEL5, TAGAP, SF1, BTG1, SLC 2A3, NCL, DDX17, ITSN2, MGAT4A, NDUFA11, GADD45GIP1, LSM5, STX7, HOOK2, TCEAL4, CLEC2D, PSMC1, ARGLU1, BCKDHA.

[0078] Protective T-cell differentially expressed genes: CST7, CTSW, HLA-DPB1, KLRG1, MYO1F, SAMD3, SRGN, GNG2, ITGAL, ATP2B4, CLIC1, ACTN4, GZMM, ACTN1, MYO1G, MSN, LCP1, CYBA, LITAF, CD8B, PYHIN1, PPP2R5C, AIF1, DBI, TLN1, PASK, LYAR, IL10RA, ALOX5AP, SASH3, ARPC5L, KLRB1 , CFLAR, RASSF1, AKNA, RAB37, ADAM8, STK10, REEP5, CTSD, TAP1, SYTL3, NR4A2, HERPUD1, DNAJC1, MFSD10, ANXA5, LMNA, IKZF3, DIP2A, G6PD, DERL1, ZBTB1, EPS15, SH2D1A, AGTRAP, PTPN4, ARHGAP25, IDI1, TFDP2, AGTPBP1, BATF, TACC1, CD84.

[0079] S500, based on AD-sensitive metabolic pathways, extracts highly expressed genes from the first transcriptome data as a feature set.

[0080] In one possible implementation, step S500 includes the following steps: S510 is based on a predefined set of genes related to nucleotide metabolism in the nucleotide metabolism pathway to identify macrophage features associated with nucleotide metabolism; and based on a predefined set of genes related to amino acid metabolism in the amino acid metabolism pathway to identify protective T / B features.

[0081] Furthermore, the set of genes related to nucleotide metabolism includes genes related to DNA synthesis and RNA synthesis.

[0082] Furthermore, the set of genes related to amino acid metabolism includes genes related to protein synthesis and degradation.

[0083] It should be noted that the aforementioned nucleotide metabolism-associated macrophage (Risk-M) characteristics refer to differentially expressed genes in monocytes that are involved in dangerous nucleotide metabolism and immune response, including: RNASET2, ADAR, AIF1, HLA-E, SERPINA1, FGR, PAPSS2, FCER1G, CSF1R, GPX1, and PTPN6.

[0084] The protective T / B characteristics refer to differentially expressed genes for B cell protection based on amino acid metabolism, including: EEF1A1, EEF1B2, EEF1G, CD74, TPT1, FOS, LAPTM5, DDX5, CD69, TOMM7, IL4R, TSC22D3, MTSS1, MEF2C, NME1-NME2, STK17B, ADK, TCL1A, UQCRB, BACH2, MYADM, CD48, CD83, EIF3E, ACTG1, FBL, FCER2, LBH, TXN, JUN. B, PPA1, PTPRC, BCLAF1, ZFP36, AIM2, SMS, PSME2, ​​TUBA1A, ZFP36L1, SLC11A1, C1QBP, KLF6, BRD2, FGR, YPEL5, TAGAP, SF1, BT G1, SLC2A3, NCL, DDX17, ITSN2, MGAT4A, NDUFA11, GADD45GIP1, LSM5, STX7, HOOK2, TCEAL4, CLEC2D, PSMC1, ARGLU1, BCKDHA.

[0085] Differentially expressed genes protecting T cells from amino acid metabolism include: CST7, CTSW, HLA-DPB1, KLRG1, MYO1F, SAMD3, SRGN, GNG2, ITGAL, ATP2B4, CLIC1, ACTN4, GZMM, ACTN1, MYO1G, MSN, LCP1, CYBA, LITAF, CD8B, PYHIN1, PPP2R5C, AIF1, DBI, TLN1, PASK, LYAR, IL10RA, ALOX5AP, SASH3, ARPC5L, K LRB1, CFLAR, RASSF1, AKNA, RAB37, ADAM8, STK10, REEP5, CTSD, TAP1, SYTL3, NR4A2, HERPUD1, DNAJC1, MFSD10, ANXA5, LMN A, IKZF3, DIP2A, G6PD, DERL1, ZBTB1, EPS15, SH2D1A, AGTRAP, PTPN4, ARHGAP25, IDI1, TFDP2, AGTPBP1, BATF, TACC1, CD84.

[0086] The functional status, subset composition, or molecular expression characteristics of T cells and B cells are defined by their core markers: "enhancing the body's resistance to pathogens (such as viruses and bacteria) or diseases (such as tumors), reducing the risk of infection / morbidity, or improving prognosis." They are widely used in immunology, infectious disease, and oncology research.

[0087] S520 extracts specific features from monocytes, T cells, and B cells and performs differential analysis. Based on the analysis results, significantly differentially expressed genes belonging to different cell models of the target metabolic pathway are screened and merged into the final feature set.

[0088] Furthermore, methods for extracting specific features include: (1) Monocytes (Risk-M model): (1.1) Extract the expression subset of monocyte samples from the first single-cell data, calculate the 75th percentile of each gene in all monocyte samples as the high expression threshold; and screen out genes with expression values ​​higher than this threshold as high expression genes. (1.2) Take the intersection of highly expressed genes and nucleotide metabolic pathways to obtain the initial feature set of Risk-M; (2) T cells (protective T model): (2.1) Extract expression subsets from T cell samples and calculate the 75th percentile threshold for each gene; screen for highly expressed genes; (2.2) Take the intersection of highly expressed genes and amino acid metabolism pathways to obtain the initial feature set of protective T; (3) B cells (protective B model): (3.1) Extract the expression subset of B cell samples and calculate the 75th percentile threshold; screen for highly expressed genes; (3.2) Take the intersection of the highly expressed genes and the amino acid metabolism pathway to obtain the initial feature set of protective B.

[0089] S530, Differential Expression Analysis Furthermore, the analysis method is as follows: Risk-M discrepancy analysis: Perform discrepancy analysis on the initial Risk-M feature set in the complete sample; Difference analysis of protective T: Perform difference analysis on the initial feature set of protective T; Difference analysis of protective B: Perform difference analysis on the initial feature set of protective B.

[0090] S540 extracts significantly differentially expressed genes and genes highly expressed in corresponding cells from the differential expression analysis results; and merges significantly differentially expressed genes from different cell models belonging to the target metabolic pathway into the final feature set.

[0091] Furthermore, the significant difference in expression met the following criteria: |log2 Fold Change|>0.5 (equivalent to a 1.5-fold change in expression), and the adjusted p-value (adj.P.Val)<0.05 (BH correction).

[0092] Specifically, the final feature set generates a txt file containing the following three types of features: Risk-M characteristics: Nucleotide metabolism genes that are highly expressed in monocytes and significantly associated with AD; Protective T characteristics: amino acid metabolism genes that are highly expressed in T cells and are significantly associated with AD protection; Protective B characteristics: amino acid metabolism genes that are highly expressed in B cells and are significantly associated with AD protection.

[0093] S600 uses the first transcription data as training data and divides it into a training set and a validation set.

[0094] For example, the sets are randomly divided into a training set (70%) and an independent validation set (30%) in a 7:3 ratio.

[0095] S700 uses AD-related genes from the S500 feature set of risk nucleotide metabolism-related macrophages (Risk-M), protective amino acid metabolism T cells (Protective-T), and protective amino acid metabolism B cells (Protective-B) as factors; and uses the training data obtained from S600 to train various machine learning models, resulting in three final training models (features such as Risk-M, Protective-T, and Protective-B).

[0096] Furthermore, the training data was randomly divided into a training set (70%) and an independent validation set (30%) in a 7:3 ratio. The expression matrix of multiple datasets (containing the expression profile of common genes) was used as input, and the binary classification labels of AD patients and healthy controls were used as output to train the model and obtain multiple training models.

[0097] Optionally, the machine learning model includes models such as random forest, XGBoost, logistic regression, support vector machine, gradient boosting machine, decision tree, k-nearest neighbor algorithm, naive Bayes, linear regression, ridge regression, lasso regression, support vector regression, decision tree regression, random forest regression, and neural network. The model's generalization ability is improved through hyperparameter optimization and cross-validation. The hyperparameters are optimized in the training set through 10-fold cross-validation (e.g., ntree=500, mtry=sqrt(n_features) for random forest).

[0098] S800 evaluates the performance of multiple trained models and selects the optimal model.

[0099] In one possible implementation, step S800 includes the following steps: S810, Cross-validation and Performance Metrics Calculation: 5-fold cross-validation is used to evaluate model stability, and the average AUC (area under the ROC curve), accuracy, sensitivity, specificity, and F1-score are calculated.

[0100] S820, using ROC curves to compare the discriminative power of different models on the validation set, selected the model with the highest AUC for further analysis. Ultimately, the optimal model was identified with seven key genes: Uqcrb, Klf6, Txn, Lsm5, Psmc1, Eif3e, and Laptm5. These key genes can be used for the prediction of Alzheimer's disease (AD).

[0101] S830, Calibration Proficiency Assessment: Plot calibration curves, calculate Brier scores and Hosmer-Lemeshow p-values, and assess the consistency between model-predicted probabilities and actual disease risks.

[0102] S840, Clinical Validation: The net benefit of the model is calculated through decision curve analysis (DCA) to quantify the clinical decision-making utility of the model under different risk thresholds.

[0103] Furthermore, methods for evaluating model performance through ROC curve, calibration curve, and decision curve analysis (DCA) include: 1. ROC curve (Receiving Operating Characteristic curve) Evaluation objective: The model's discriminative ability (i.e., the accuracy in distinguishing AD patients from healthy controls).

[0104] Steps: Curve generation: Iterate through all possible classification thresholds (0-1) and calculate the true positive rate (TPR) and false positive rate (FPR) for each threshold. Plot a curve with FPR on the horizontal axis and TPR on the vertical axis.

[0105] Key metric: AUC (area under the curve): range 0.5 (random guess) to 1 (perfect classification), AUC>0.8 indicates that the model has high discriminative ability.

[0106] Optimal threshold selection: The optimal threshold is determined by maximizing the Youden exponent (TPR + TNR -1), balancing sensitivity and specificity.

[0107] Interpretation: The higher the AUC, the stronger the model's discriminative ability; if the AUC is close to 0.5, the model has no practical predictive value.

[0108] 2. Calibration Curve Evaluation objective: The calibration of the model's predicted probabilities (i.e., the consistency between predicted risk and actual risk).

[0109] Steps: Curve generation: Divide the predicted probabilities into bins (e.g., 10 groups) and calculate the average predicted probability and actual prevalence for each group.

[0110] Plot a curve with the predicted probability on the horizontal axis and the actual probability on the vertical axis; ideally, it would be a straight line at a 45-degree angle.

[0111] Key metric: Brier score: range 0 (perfect calibration) ~ 0.25 (complete error), the smaller the value, the higher the calibration accuracy.

[0112] Hosmer-Lemeshow test: p-value > 0.05 indicates that there is no significant difference between the predicted and actual values ​​(good calibration).

[0113] Interpretation: If the calibration curve deviates from the diagonal (such as an S-shape or U-shape), it indicates that the model overestimates or underestimates the risk and needs to be recalibrated using Platt Scaling or Isotonic Regression.

[0114] 3. Decision Curve Analysis (DCA) Evaluation objective: Clinical applicability of the model (i.e., net benefit at different risk thresholds).

[0115] Steps: Curve generation: Set a risk threshold range (e.g., 0%~50%), and calculate the net benefit of the model at each threshold: Plot the net benefit as a function of the threshold and compare it with the "full intervention" (assuming all samples are positive) and "no intervention" (assuming all samples are negative) strategies.

[0116] Key Metric: Net Benefit Value: The higher the value, the higher the clinical decision-making utility of the model at a specific threshold.

[0117] Threshold probability range: The wider the threshold range covered by the model curve, the stronger its clinical applicability.

[0118] Interpretation: If the model curve is higher than the "full intervention" and "no intervention" lines in a certain threshold range, then using the model in that range can bring additional clinical benefits.

[0119] Model interpretability analysis: SHAP value calculation: Quantifying the contribution of each gene using Python's shap package: Laptm5 and Eif3e show positive contributions (SHAP>0), while Klf6 and Psmc1 show negative contributions (SHAP<0).

[0120] 4. Comprehensive Performance Evaluation Strategy Priority ranking: Discrimination ability (AUC) > Calibration (Brier score) > Net clinical benefit (DCA curve coverage).

[0121] Models with high AUC are prioritized for retention, and then models with good calibration and strong clinical applicability are selected from them.

[0122] 5. Evaluate the effectiveness of constructing AD prediction models in differentiating other common aging and metabolic disorders (Parkinson's disease, frontotemporal dementia, and obesity). The first transcriptome data was integrated with control data from patients with Parkinson's disease, frontotemporal dementia, and obesity to obtain the evaluation dataset.

[0123] Specifically, Parkinson's disease (PD) data: GSE6613, GSE57475, GSE72267, GSE99039.

[0124] Frontotemporal dementia (FTD) data: GSE140829, GSE140830.

[0125] Obesity (OB) data: GSE109597, GSE110551, GSE181339.

[0126] The data was downloaded and parsed using the GEOquery package in R, and stored as an expression matrix and sample metadata.

[0127] The effectiveness of the optimal model in differentiating common aging and metabolic disorders was evaluated using an evaluation dataset.

[0128] S900, based on an optimal model, constructs a dynamic nomogram containing key genes to predict the risk of AD in individual patients.

[0129] In one possible implementation, step S900 includes the following steps: S910, Data Preparation: Extract nucleotide metabolism-related macrophage and protective T / B cell features from the training set, converting the extracted feature genes into factor types.

[0130] S920 uses a generalized linear model to perform classification analysis on the transformed training set to obtain a logistic regression model.

[0131] Furthermore, the generalized linear model (logistic regression): a generalized linear model is constructed using the glm function, with the model formula being status ~ gene1 + gene2 + gene3. The data source is the training set, and the binomial distribution family (binomial()) is used.

[0132] S930, Model Visualization: Use the regplot function to visualize the results of the logistic regression model.

[0133] For example, in Figure 4 Image 'a' shows the details of the second observation in the lung dataset. The parameters are set as follows: points = T to display points, center = T to prevent variable points from starting at 0, odds = F to suppress the OR hazard ratio, showP = T to display the statistical significance of each variable, rank = 'sd' to sort by the SD values ​​of the regression coefficients, clickable = F to prevent further image editing, title = 'Nomgram' to set the title name, and interval = 'confidence' to plot the 95% confidence interval.

[0134] S940, Creating a Dynamic Nomogram: Use the DynNom function to create a dynamic nomogram to display the prediction results of the logistic regression model. Set DNtitle to 'Nomogram' and DNxlab to 'Probability'.

[0135] S950 inputs gene expression values ​​into a dynamic nomogram to assess the risk of AD in the subject.

[0136] The technical solution of the present invention will be further described below with reference to more specific embodiments.

[0137] 1. Prediction Methods 1.1 Data Acquisition and Preprocessing The GWAS data for Alzheimer's disease (AD) patients came from the MRC-IEU GWAS Summary Dataset (MRC Integrative Epidemiology Unit, OpenGWAS project). This dataset is a database containing genetic association information, encompassing genetic association information from the GWAS Summary Dataset, and can be queried or downloaded. This dataset contains 488,285 samples and 12,321,875 individual SNPs (single nucleotide polymorphisms), providing rich genetic information for research.

[0138] To further enrich the genetic map, the study also obtained aggregated statistics of 731 GWAS related to immune cells from the GWAS Catalog. These data, with GWAS IDs ranging from GCST90001391 to GCST90002121, cover a wide range of immune-related genetic variations, helping to reveal the potential role of the immune system in the pathogenesis of Alzheimer's disease (AD).

[0139] In addition, the study downloaded data from a large-scale plasma metabolite GWAS cohort from peer-reviewed publications. This cohort included 8,299 participants, 1,091 metabolites, and 309 metabolite ratios, providing important data support for studying the relationship between metabolites and AD.

[0140] It is worth noting that all three datasets are derived from European populations. This consistency in population origin ensures the homogeneity of the research analysis, thereby improving the reliability and comparability of the results. By integrating these multidimensional genetic and metabolic data, the study can more comprehensively explore the genetic basis and potential biological mechanisms of Alzheimer's disease (AD), providing a scientific basis for the prevention, diagnosis, and treatment of AD.

[0141] Transcriptome data: Blood single-cell transcriptome data from AD patients in Europe and China (GSE63060, GSE63061, and the Chinese cohort) were integrated and preprocessed to obtain the first transcriptome data. In addition, control data from patients with Parkinson's disease (PD), frontotemporal dementia (FTD), and obesity (OB) were introduced to validate the model against rejection.

[0142] To delve deeper into the molecular mechanisms and potential biomarkers of Alzheimer's disease (AD), we obtained transcriptomic data from various AD patients from the Gene Expression Omnibus (GEO) database. These datasets encompass different types of samples, including peripheral blood, single-cell data, and brain tissue samples, providing multidimensional information support for our research.

[0143] (1) GSE63060: Peripheral blood samples from European AD patients: Sample composition: including 104 healthy controls and 145 AD patients.

[0144] (2) GSE63061: Peripheral blood samples from European AD patients: Sample composition: including 135 healthy controls and 139 AD patients.

[0145] Significance of the study: Similar to GSE63060, this dataset further expands the scope of peripheral blood sample studies, providing independent data support for validating and supplementing the results of GSE63060.

[0146] (3) Chinese cohort: AD group: 44 patients diagnosed by PET (meeting the NINCDS-ADRDA criteria), from Renmin Hospital of Wuhan University, Zhongnan Hospital and Henan Provincial People's Hospital. Healthy control group: 30 healthy individuals (excluding AD and other dementias), from community recruitment and patients visiting the same hospital.

[0147] (4) Parkinson's disease (PD) related dataset GSE6613: Includes 55 healthy controls and 50 Parkinson's disease patients.

[0148] GSE57475: Includes 49 healthy controls and 93 Parkinson's disease patients.

[0149] GSE72267: Includes 19 healthy controls and 40 Parkinson's disease patients.

[0150] GSE99039: Includes 89 healthy controls and 132 Parkinson's disease patients.

[0151] Significance of the study: These datasets provide peripheral blood transcriptome data from PD patients, which helps to study the commonalities and differences between AD and PD, and to explore the common mechanisms of neurodegenerative diseases.

[0152] (5) Frontotemporal dementia (FTD) related datasets GSE140829: Includes 249 healthy controls and 204 patients with FTD.

[0153] GSE140830: Includes 139 healthy controls and 139 patients with FTD.

[0154] Significance of the study: These datasets provide peripheral blood transcriptome information of patients with FTD, which helps to study the similarities and differences in the molecular mechanisms of FTD and AD, and provides a basis for differential diagnosis.

[0155] (6) Data of patients with a body mass index greater than 30 (OB) GSE109597: Includes 60 healthy controls and 2 patients with FTD.

[0156] Significance of the study: This dataset provides blood transcriptome data from obese patients, which helps to study the potential link between obesity and neurodegenerative diseases and explore the role of metabolic pathways in the diseases.

[0157] By integrating these diverse datasets, the study was able to explore the molecular mechanisms of Alzheimer's disease (AD) from different perspectives, including peripheral blood biomarkers, cellular heterogeneity at the single-cell level, pathological changes in brain tissue, and comparative analyses with other neurodegenerative diseases. These data provide an important foundation for the early diagnosis, disease monitoring, and development of treatment strategies for AD.

[0158] 1.2 Model Construction and Optimization 1.2.1 Feature Selection: Based on AD-sensitive metabolic pathways, a method for extracting highly expressed genes from the first transcriptome data as a feature set includes: Differentially expressed genes in monocytes related to dangerous nucleotide metabolism and immune response: RNASET2, ADAR, AIF1, HLA-E, SERPINA1, FGR, PAPSS2, FCER1G, CSF1R, GPX1, PTPN6.

[0159] Differentially expressed genes for Protective-B cells: EEF1A1, EEF1B2, EEF1G, CD74, TPT1, FOS, LAPTM5, DDX5, CD69, TOMM7, IL4R, TSC22D3, MTSS1, MEF2C, NME1-NME2, STK17B, ADK, TCL1A, UQCRB, BACH2, MYADM, CD48, CD83, EIF3E, ACTG1, FBL, FCER2, LBH, TXN, JUNB, PPA 1. PTPRC, BCLAF1, ZFP36, AIM2, SMS, PSME2, ​​TUBA1A, ZFP36L1, SLC11A1, C1QBP, KLF6, BRD2, FGR, YPEL5, TAGAP, SF1, BTG1 , SLC2A3, NCL, DDX17, ITSN2, MGAT4A, NDUFA11, GADD45GIP1, LSM5, STX7, HOOK2, TCEAL4, CLEC2D, PSMC1, ARGLU1, BCKDHA.

[0160] Differentially expressed genes for protective T cells: CST7, CTSW, HLA-DPB1, KLRG1, MYO1F, SAMD3, SRGN, GNG2, ITGAL, ATP2B4, CLIC1, ACTN4, GZMM, ACTN1, MYO1G, MSN, LCP1, CYBA, LITAF, CD8B, PYHIN1, PPP2R5C, AIF1, DBI, TLN1, PASK, LYAR, IL10RA, ALOX5AP, SASH3, ARPC5 L, KLRB1, CFLAR, RASSF1, AKNA, RAB37, ADAM8, STK10, REEP5, CTSD, TAP1, SYTL3, NR4A2, HERPUD1, DNAJC1, MFSD10, ANXA5, LM NA, IKZF3, DIP2A, G6PD, DERL1, ZBTB1, EPS15, SH2D1A, AGTRAP, PTPN4, ARHGAP25, IDI1, TFDP2, AGTPBP1, BATF, TACC1, CD84.

[0161] 1.2.2 Machine Learning Training: Based on the previously selected set of metabolic pathway-related genes (such as nucleotide metabolism-related genes and amino acid metabolism-related genes), the expression levels of the corresponding genes in the integrated datasets (GSE63060, GSE63061, and the Chinese cohort) were extracted as feature inputs. A total of 113 machine learning algorithms, including Random Forest, XGBoost, Logistic Regression, and Support Vector Machine, were used to perform binary classification modeling for AD and normal samples. Each algorithm was trained and validated based on the aforementioned gene features to systematically evaluate the impact of different feature combinations and modeling strategies on AD classification performance. Performance was evaluated using ROC curves, calibration curves, and decision curve analysis (DCA).

[0162] The results and analysis are as follows: The immune metabolic status of peripheral blood mononuclear cells (PBMCs) can help in constructing AD prediction models with cross-population effectiveness. Since single nucleotide polymorphism (SNP)-dependent immunometabolic features can modulate AD susceptibility and alter the metabolic and immune response status of PBMCs, these features can be considered helpful in establishing stable and reliable AD prediction models. In this application, cross-analysis was performed on differentially expressed genes from blood transcriptomes (AD group vs. normal control group) and immune cells (high-metabolic group vs. low-metabolic group), obtaining three differentially expressed genes for machine learning training targeting monocytes, T cells, and B cells. Figure 1 a), labeled as risk monocytes (Risk-M, high nucleotide metabolism monocytes), protective T cells (Protective-T, high amino acid metabolism T cells), and protective B cells (Protective-B, high amino acid metabolism B cells), respectively.

[0163] By integrating the blood transcriptome datasets from GSE63060, GSE63061, and the Chinese cohort (all containing AD and normal control samples), the data were randomly divided into 7 / 10 training sets and 3 / 10 validation sets. Figure 1 b). 113 machine learning algorithms were applied (see Table 1). 7 / 10 of the data was used as the training set to determine the optimal prediction conditions, and the remaining 3 / 10 was used as the independent validation set. Figure 1 c). The prediction model based on differentially expressed genes in protective B cells, obtained using the Enet[alpha=0.6]+RF machine learning method (referred to as the Protective-B model), performed best, with AUC values ​​exceeding 0.828 on both the training and validation sets. Figure 1 d and 1e). Decision curve analysis (DCA) and calibration curves further confirm that the model has significant clinical applicability and predictive power. Figure 1 f and 1g).

[0164] Table 1 113 Machine Learning Models

[0165] The Protective-B model contains seven core genes: Uqcrb, Klf6, Txn, Lsm5, Psmc1, Eif3e, and Laptm5. These genes show differential expression in the blood transcriptomes of AD patients and normal controls. Figure 1 h). SHAP analysis revealed that Laptm5 and Klf6 had bidirectional regulatory effects (SHAP values ​​ranging from approximately -0.2 to 0.2), potentially enhancing or weakening the predictive results; Txn and Lsm5 mainly exerted positive effects, while Eif3e, Uqcrb, and Psmc1 had weaker effects. Figure 1 i). The force-directed plot shows the feature contribution of individual samples (predicted value 0.102 vs. baseline 0.556): Laptm5 (11.4) and Eif3e (11.1) significantly enhanced the prediction, Klf6 (9.9) and Psmc1 (9.84) decreased the prediction, and Txn (8.33) showed a weak negative effect. Figure 1 (j) In summary, the blood transcriptome-based AD prediction model demonstrates high-precision predictive ability that can be clinically translated and robust cross-population effectiveness. It should be noted that Uqcrb, also known as ubiquinol-cytochrome c reductase binding protein, is located on human chromosome 10, specifically at 10q23.33. This gene encodes a subunit of the ubiquinol-cytochrome c oxidoreductase complex, which consists of a mitochondrial-encoded subunit and 10 nuclear-encoded subunits.

[0166] Klf6, short for Kruppel-like factor 6, is located on human chromosome 10p15. This gene encodes a protein belonging to the Zinc fingers C2H2-type family, a class of proteins that play a crucial role in transcriptional regulation.

[0167] TXN, also known as thioredoxin, is located on human chromosome 9q31.3. The protein encoded by this gene plays a crucial role in cells, particularly in redox reactions. The product of the TXN gene, thioredoxin, is a small protein that regulates redox states within cells, influencing a variety of biological processes.

[0168] Lsm5, short for Like Sm domain-containing protein 5, is an important member of the Lsm protein family (Sm-like proteins family). Located on chromosome 7p14.3, its core functions are closely related to the processing, metabolism, and stability regulation of RNA, and it plays a key role in normal cellular physiological activities and disease development.

[0169] Psmc1, short for proteasome 26S subunit, ATPase 1, is a protein-coding gene located on chromosome 19q13.2. It belongs to the AAA ATPases, Proteasome family, whose members are all closely related to protein degradation. PSMC1 is also known as S4, p56, and RPT2, names reflecting its specific role in the protein degradation complex.

[0170] Eif3e, also known as eIF3-p48, is a component of the eukaryotic translation initiation factor 3 (eIF3) complex, located at position 16p13.3 on human chromosome 16. This complex plays a crucial role in protein synthesis, particularly in the translation initiation phase.

[0171] Laptm5, short for lysosomal protein transmembrane 5, is located on human chromosome 1p36.33. This gene encodes a protein belonging to the microRNA protein-coding host gene family. In biological databases, its OMIM ID is 601476, its ENSEMBLE ID is ENSG00000162511, and its UniProt ID is Q13571.

[0172] Based on the above results, the Protective-B model can even predict the risk of familial Alzheimer's disease before symptom detection.

[0173] Although the Alzheimer's disease (AD) prediction clues in this study are based on single nucleotide polymorphism (SNP)-dependent immunometabolic features in sporadic AD patients, the data strongly suggest that infection and immune response are involved in AD pathogenesis, which is also considered one of the contributing factors to familial AD. In this context, the Protective-B model may also be applicable to the prediction of familial AD. Therefore, this application example uses a cohort of 9-month-old wild-type (WT) and 5xFAD transgenic mice for validation experiments. Before collecting blood samples, a water maze test was performed according to the methods described in the literature to assess learning and memory abilities (…). Figure 2 The results showed that although 9-month-old 5xFAD mice had relatively fast movement speed, their escape latency was significantly prolonged, the number of times they crossed the target area was reduced, and the time they spent in the target area was slightly shortened, suggesting that they had learning and memory dysfunction. Figure 2 bf). Immunofluorescence staining confirmed significant amyloid protein deposition in the brains of 9-month-old 5xFAD mice. Figure 2 g). Inputting transcriptome data into the prediction model yielded an area under the curve (AUC) of 1.000, significantly higher than the prediction performance of a single gene. Figure 2 h, j). To verify the predictive ability of this model for early AD, this application further tested 2.5-month-old 5xFAD mice. Mice in this age group showed only mild memory impairment and limited amyloid deposition (h, j). Figure 2 (ko), but the predictive model still maintains good performance (AUC=0.922) Figure 2 These results indicate that the model can effectively predict the progression of familial Alzheimer's disease, even in the preclinical stage before the onset of symptoms.

[0174] The Protective-B model is specific and unaffected by metabolic diseases.

[0175] To assess model specificity, embodiments of this application further tested its ability to differentiate other neurodegenerative diseases with overlapping clinical phenotypes, particularly Parkinson's disease [PD] and frontotemporal dementia [FTD]. After standardized preprocessing, receiver operating characteristic (ROC) curve analysis of peripheral blood transcriptome data from PD and FTD patients showed that the Protective-B model had an AUC value of 0.814 in AD patients. Figure 3 a), while in PD and FTD patients it was only 0.637 ( Figure 3 b) and 0.603 ( Figure 3 c). It is worth noting that when applied to obese individuals (OB, body mass index > 30), the AUC value was 0.644 ( Figure 3d) There is no overlap with the AD population, confirming that metabolic abnormalities such as obesity do not interfere with AD prediction. Further analysis shows that the upper limit of the 95% confidence interval (CI) of the ROC curves for the three groups (PD, FTD, OB) did not exceed 0.71, which is still lower than the lower limit of the 95% CI for the AD group (0.763), indicating that the model has accurate AD identification capabilities. In summary, the AD prediction model described in this application has high disease specificity and is unaffected by metabolic diseases, significantly improving the reliability of clinical applications.

[0176] 1.4 Development of Dynamic Nodal Charts: Optimized dynamic nodal charts for personalized AD prediction and their clinical translational applications.

[0177] To facilitate clinical application, this application's embodiments construct a nomogram based on seven independent predictors selected by the Protective-B machine learning model. Figure 4 a). This tool assesses AD risk by assigning scores to each factor and projecting the total score onto a risk axis. The online version of the dynamic nomogram, developed to improve clinical usability, showed statistical significance (P<0.05). Figure 4 b). By including only statistically significant factors—Txn, Laptm5, Psmc1, and Klf6 (P < 0.05, Figure 4 cd), the optimized nomogram finally obtained an AUC value of 0.802 ( Figure 4 e). Among them, the Klf6 gene alone showed the best predictive power (AUC=0.714, Figure 4 f). Both decision curve analysis and calibration curves confirm that the model has high reliability. Figure 4 The final deployed online AD risk prediction tool (https: / / jw-ad.shinyapps.io / DynNomapp / ) allows users to assess risk by inputting gene expression levels. Figure 4 (The example shown is a cohort test of a Chinese population). This nomogram allows users to input individualized gene expression data, providing clinicians and researchers with an interactive interface for quantitative AD risk assessment based on specific gene expression characteristics.

[0178] A dynamic nomogram containing key genes such as Txn, Laptm5, Psmc1, and Klf6 was constructed (Web tool link: https: / / jiwu96.shinyapps.io / AD_DynNomapp / ), allowing users to input gene expression values ​​for rapid assessment of AD risk.

[0179] The prediction methods have been explained above. The following section introduces products for diagnosing Alzheimer's disease.

[0180] This application provides an application of a reagent for detecting a gene combination in the preparation of products for diagnosing Alzheimer's disease, wherein the gene combination includes: Uqcrb, Klf6, Txn, Lsm5, Psmc1, Eif3e, and Laptm5.

[0181] Based on the aforementioned prediction method, the gene combination selected in this application has a good predictive effect. Medical personnel can combine the genes together to detect and diagnose the sample to be tested, in order to diagnose whether the sample to be tested has AD.

[0182] This application also provides a kit that may contain the detection reagents involved in the aforementioned prediction method to diagnose whether a sample to be tested has Alzheimer's disease (AD). The limitations and technical solutions of the kit in this application can be found in the description of the aforementioned prediction method, and will not be repeated here. Similarly, the above-mentioned kit can also be used in the preparation of AD detection products, and will not be repeated here.

[0183] The following describes the integrated multi-omics anatomical Alzheimer's disease prediction device based on immunometabolic crosstalk provided by the present invention. The integrated multi-omics anatomical Alzheimer's disease prediction device based on immunometabolic crosstalk described below can be referred to in correspondence with the integrated multi-omics anatomical Alzheimer's disease prediction method based on immunometabolic crosstalk described above.

[0184] Figure 5 This is a schematic diagram of the structure of the integrated multi-omics anatomical Alzheimer's disease prediction device based on immune metabolic crosstalk provided in an embodiment of the present invention, as shown below. Figure 5 As shown, it includes: acquisition module 51, preprocessing module 52, analysis module 53, gene screening module 54, segmentation module 55, extraction module 56, training module 57, model screening module 58, and prediction module 59, wherein: Acquisition module 51 is used to acquire transcriptome data and single-cell data from AD patients and healthy individuals; Preprocessing module 52 is used to preprocess transcriptome data and single-cell data respectively to obtain first transcriptome data and first single-cell data; Analysis module 53 is used to perform clustering, dimensionality reduction, and pathway enrichment analysis on the first single-cell data to identify nucleic acid metabolism pathways in glial cells and monocytes, as well as amino acid metabolism pathways in neurons and T / B cells; differential gene analysis is performed to obtain gene analysis results; The gene screening module 54 is used to analyze the expression differences of different metabolic pathways in various cell populations based on the positive results obtained from gene analysis; and to screen out genes related to AD disease. The partitioning module 55 is used to divide the first transcribed data into a training set and a validation set, using the first transcribed data as training data. Extraction module 56 is used to extract highly expressed genes from the first transcriptome data as a feature set based on AD-sensitive metabolic pathways; Training module 57 is used to use AD-related genes of macrophages, protective T cells, and protective B cells related to nucleotide metabolism as factors; and to train multiple machine learning models using training data to obtain three final training models. The model selection module 58 is used to evaluate the performance of multiple trained models and select the optimal model. Prediction module 59 is used to construct a dynamic nomogram containing key genes based on the optimal model; and to predict the risk of AD in a single patient.

[0185] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device may include a processor 610, a communications interface 620, a memory 630, and a communication bus 640. The processor 610, communications interface 620, and memory 630 communicate with each other via the communication bus 640. The processor 610 can call logical instructions from the memory 630 to execute a comprehensive multi-omics anatomical Alzheimer's disease prediction method based on immune metabolic crosstalk.

[0186] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0187] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the comprehensive multi-omics anatomical Alzheimer's disease prediction method based on immune metabolic crosstalk provided by the above methods.

[0188] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the integrated multi-omics anatomical Alzheimer's disease prediction method for immune metabolic crosstalk provided by the methods described above.

[0189] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0190] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. Use of a reagent for detecting a combination of genes in the manufacture of a diagnostic product for Alzheimer's disease, said combination of genes comprising: Uqcrb, Klf6, Txn, Lsm5, Psmc1, Eif3e and Laptm5. ​ 2. A kit characterized in that, The kit comprises detection reagents for detecting a combination of genes, the combination of genes comprising: Uqcrb, Klf6, Txn, Lsm5, Psmc1, Eif3e and Laptm5.

3. The comprehensive multi-omics dissection of immunometabolic string for Alzheimer's disease prediction method characterized in that, The method comprises the following steps: Obtaining the transcriptome data and single-cell data of AD patients and healthy people; Preprocessing the transcriptome data and single-cell data to obtain first transcriptome data and first single-cell data; Clustering, dimensionality reduction and pathway enrichment analysis are performed on the first single-cell data to identify nucleic acid metabolism pathways in glial cells and monocytes, and amino acid metabolism pathways in neurons, T / B cells; Differential gene analysis is performed to obtain gene analysis results; According to the positive results obtained by gene analysis, the expression differences of different metabolic pathways in various cell populations are analyzed; and genes related to AD disease are screened out; The first transcript data is divided into a training set and a validation set as training data; Based on the AD-sensitive metabolic pathways, high-expression genes in the first transcriptome data are extracted as a feature set; The AD-related genes of nucleotide metabolism-related macrophages, protective T cells and protective B cells in the feature set are used as factors; A plurality of machine learning models are trained using the training data to obtain three final training models; The model performance of the plurality of training models is evaluated, and the optimal model is selected; Based on the optimal model, a dynamic column diagram containing key genes is constructed; and the AD risk of an individual patient is predicted.

4. The integrated multi-omic anatomical Alzheimer’s disease prediction method of immunometabolic cross-talk according to claim 3, characterized in that, Clustering, dimensionality reduction and pathway enrichment analysis are performed on the first single-cell data to identify nucleic acid metabolism pathways in glial cells and monocytes, and amino acid metabolism pathways in neurons, T / B cells; Differential gene analysis is performed to obtain gene analysis results, including: According to the clustering analysis results, cell clustering and annotation are performed to identify cell populations that have an important impact on the occurrence of AD; Cell population subclustering and gene expression level analysis: the cell population is subclustered; and the gene expression levels of metabolite pathways in different cell populations are analyzed; The T cell, B cell and monocyte cell populations are extracted from the subclustered cell populations; and according to the gene expression levels of the metabolite pathways, the cells are divided into an immune cell high-metabolism group and a low-metabolism group; Differential gene analysis is performed to obtain gene analysis results.

5. The comprehensive multi-omic anatomical Alzheimer’s disease prediction method of immunometabolic cross-talk according to claim 3, characterized in that, Based on the AD-sensitive metabolic pathways, high-expression genes in the first transcriptome data are extracted as a feature set, including: Based on the nucleotide metabolism pathway, a gene set related to nucleotide metabolism is predefined to identify nucleotide metabolism-related macrophage characteristics; and based on the amino acid metabolism pathway, a gene set related to amino acid metabolism is predefined to identify protective T / B characteristics; Specific feature extraction is performed on monocytes, T cells and B cells, and differential analysis is performed; based on the analysis results, significant differential genes of different cell models belonging to the target metabolic pathway are screened out and combined into a final feature set; Differential expression analysis; Extract significant differentially expressed and significantly high expression genes in corresponding cells from the differential expression analysis results; combine the significant differentially expressed genes of different cell models belonging to the target metabolic pathway into the final feature set.

6. The integrated multi-omics dissection of Alzheimer's disease prediction method of immunometabolic string according to claim 1, characterized in that, The method for predicting the AD risk of a single patient based on the optimal model includes constructing a dynamic column diagram containing key genes. The training set is subjected to nucleotide metabolism-related macrophage, protective T cell and protective B cell feature extraction, so that the extracted feature genes in the training set are converted into factor types. A generalized linear model is used to perform classification analysis on the converted training set to obtain a logistic regression model. The results of the logistic regression model are visualized. Dynamic column diagram creation is used to display the prediction results of the logistic regression model. Gene expression values are input into the dynamic column diagram to evaluate the risk of AD in the testee.

7. An integrated multi-omic dissection of immunometabolic string for Alzheimer's disease prediction apparatus, characterized in that, The method comprises the following steps: An acquisition module is configured to acquire transcriptome data and single-cell data of AD patients and healthy people. A preprocessing module is configured to preprocess the transcriptome data and single-cell data to obtain first transcriptome data and first single-cell data. An analysis module is configured to perform clustering, dimensionality reduction and pathway enrichment analysis on the first single-cell data to identify nucleic acid metabolism pathways in glial cells and monocytes, and amino acid metabolism pathways in neurons, T cells and B cells. Differential gene analysis is performed to obtain gene analysis results. A gene screening module is configured to analyze the expression differences of different metabolic pathways in various cell populations based on the positive results obtained from the gene analysis, and screen genes related to AD. A division module is configured to divide the first transcript data into a training set and a validation set as training data. An extraction module is configured to extract high expression genes in the first transcriptome data as a feature set based on AD-sensitive metabolic pathways. A training module is configured to use AD-related genes of nucleotide metabolism-related macrophages, protective T cells and protective B cells in the feature set as factors. A plurality of machine learning models are trained using the training data to obtain three final training models. A model screening module is configured to evaluate the model performance of the plurality of training models and screen the optimal model. A prediction module is configured to construct a dynamic column diagram containing key genes based on the optimal model, and predict the AD risk of a single patient.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the comprehensive multi-omics dissection of Alzheimer's disease prediction method of the immune metabolism string according to any one of claims 3 to 6. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the comprehensive multi-omics dissection of Alzheimer's disease prediction method of the immune metabolism string according to any one of claims 3 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the comprehensive multi-omics dissection of Alzheimer's disease prediction method of the immune metabolism string according to any one of claims 3 to 6.

Images