Application of GOT1, MT3, SETD1B, FH and HRAS genes and method for screening Alzheimer's disease related genes

Abstract

The invention relates to the technical field of gene screening, in particular to application of GOT1, MT3, SETD1B, FH and HRAS genes and a method for screening Alzheimer's disease related genes, and five genes related to the Alzheimer's disease, namely the GOT1, MT3, SETD1B, FH and HRAS genes, are obtained through screening. The five genes obtained by screening can be used for constructing a model for predicting the probability of the Alzheimer's disease and a drug model for predicting and treating the Alzheimer's disease, and important reference is provided for diagnosis, mechanism research and treatment of the Alzheimer's disease in the later period.

Description

Technical Field

[0001] The present invention relates to the technical field of gene screening, and in particular to the application of GOT1, MT3, SETD1B, FH and HRAS genes and a method for screening genes related to Alzheimer's disease. Background Art

[0002] Alzheimer's disease (AD), as a neurodegenerative disease, is often accompanied by cognitive dysfunction and memory impairment, and even affects language and motor skills. At present, the diagnosis of AD mainly relies on PET of tracer molecules and analysis of CSF proteins, but due to the high cost of specialized PET scans, it is not widely used, and the analysis of CSF proteins takes a long time to obtain results due to invasive methods (lumbar puncture) and lack of laboratory facilities involving fluid analysis, which affects the treatment of AD. Therefore, in order to diagnose and treat AD patients earlier, it is urgent to explore biomarkers that can diagnose AD patients at an early stage.

[0003] N6-methylladenosine (m6A) is the most common post-transcriptional modification of mRNA in mammals. It occurs at the 6th N atom of adenine. This dynamic and reversible epigenetic modification affects the fate of RNA molecules, thereby regulating processes such as cell differentiation, immunity and metabolism in mammals. m6A can play a key role in regulating mRNA modifications related to various biological processes by affecting the stability, splicing, translation and localization of mRNA. Studies have shown that m6A is associated with neurodegenerative diseases, and potential perturbations in m6A signaling may lead to AD neuronal dysfunction.

[0004] Ferroptosis is an iron-dependent, novel type of programmed cell death that is different from apoptosis, necrosis, and autophagy. Emerging evidence suggests that ferroptosis can be found in neuronal cell death associated with various neurological diseases, such as hemorrhagic stroke, ischemic stroke, Parkinson's disease, and Huntington's disease. Studies have found that ferroptosis is involved in the occurrence and development of AD. For example, excessive iron accumulates in insoluble amyloid plaques and neurofibrillary tangles, which are typical pathological features of Alzheimer's disease.

[0005] Studies have shown that m6A may be a key transcriptional regulator of ferroptosis in stellate cells in liver fibrosis, suggesting that in non-tumor diseases, m6A may interact with ferroptosis genes, thus playing a key role in the occurrence and development of the disease. Therefore, exploring the potential biomarkers of m6A and ferroptosis-related genes in AD and their possible regulatory mechanisms will provide a certain theoretical basis for the early diagnosis and drug treatment of AD. Summary of the invention

[0006] The purpose of the present invention is to develop an application of GOT1, MT3, SETD1B, FH and HRAS genes and a method for screening genes related to Alzheimer's disease.

[0007] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0008] The present invention provides the use of one or more of GOT1, MT3, SETD1B, FH and HRAS genes in preparing a drug for preventing and / or treating Alzheimer's disease.

[0009] The present invention also provides the use of one or more of the GOT1, MT3, SETD1B, FH and HRAS genes in constructing a model for predicting the probability of suffering from Alzheimer's disease.

[0010] The present invention also provides the use of one or more of the GOT1, MT3, SETD1B, FH and HRAS genes in preparing a product that can predict drugs for treating Alzheimer's disease.

[0011] The present invention also provides a method for screening genes related to Alzheimer's disease, comprising the following steps:

[0012] (1) The GSE5281 and GSE48350 datasets, which include normal samples and Alzheimer's disease samples, were merged as training sets;

[0013] (2) Genes related to N6-methyl in the training set were extracted, and the obtained N6-methyl-related genes were tested using wilcox.test. The false discovery rate (FDR) was used to test the pValue, and genes with a threshold of fdr < 0.05 were screened to obtain the N6-methyl-related gene DE-m6A differentially expressed in normal samples and Alzheimer's disease samples in the training set;

[0014] (3) Using the spearman algorithm, the correlation between the DE-m6A obtained in step (2) and ferroptosis-related genes was analyzed, and ferroptosis-related genes with correlation > 0.4 and P value < 0.05 were selected to obtain genes DE-m6A-FRG that were both related to N6-methyl and ferroptosis;

[0015] (4) The SVM-RFE and Boruta algorithms were used to screen the characteristic genes in DE-m6A-FRG respectively, and the characteristic genes with intersection under the two algorithms were selected to obtain genes related to Alzheimer's disease.

[0016] Preferably, the GSE5281 and GSE48350 datasets described in step (1) are both derived from the GEO database.

[0017] Preferably, the genes associated with N6-methyl in step (2) include METTL4, KIAA1429, RBM15, METTL5, METTL14, WTAP, METTL3, METTL16, ZC3H13, ZCCHC4, YTHDC1, YTHDF1, YTHDF2, YTHDF3, IGF2BP1, IGF2BP2, IGF2BP3, eIF3, PRRC2A, FMRP, YTHDC2, HNRNPG, HNRNPC, HNRNPA2B1, ALKBH5, FTO and RBM15B.

[0018] Preferably, the ferroptosis-related genes in step (3) include RPL8, ATP5MC3, CS, EMC2, NOX1, CYBB, NOX3, NOX5, DUOX1, DUOX2, G6PD, PGD, VDAC2, PIK3CA, FLT3, SCP2, TP53, LPCAT3, NRAS, KRAS, HRAS, TFR2, GOT1, CARS1, ATG7, NCOA4, ALOX12B, ALOX15B, ALOXE3, PHKG2, ACO1, SLC38A1, GLS2, G6PDX, ULK1, ATG3, ATG4D, ATG5, BECN1, MAP1LC3A, G ABARAPL2, GABARAPL1, ATG16L1, WIPI1, WIPI2, SNX4, ATG13, ULK2, SAT1, EGFR, MAPK3, MAPK1, BID, ZEB1, KEAP1, DPP4, ALOX15, CDKN2A, PEBP1, SOCS1, CD O1, MYB, MAPK8, MAPK9, SLC1A5, CHAC1, MAPK14, LINC00472, PRKAA2, PRKAA1, ABCC1, MIR6852, ACVR1B, TGFBR1, BAP1, EPAS1, HILPDA, HIF1A, ALOX12, ACS L4, HMOX1, IFNG, ANO6, LPIN1, HMGB1, TNFAIP3, TLR4, NOX4, ATF3, ATM, YY1AP1, EGLN2, MIOX, TAZ, MTDH, IDH1, SIRT1, FBXW7, PANX1, DNAJB6, BACH1, LONP 1. PTGS2, DUSP1, NOS2, NCF2, MT3, UBC, ALB, TXNRD1, SRXN1, GPX2, BNIP3, OXSR1, SELENOS, ANGPTL7, DDIT4, LOC284561, ASNS, TSC22D3, DDIT3, JDP2, SES N2, SLC1A4, PCK2, TXNIP, VLDLR, GPT2, PSAT1, LURAP1L, SLC7A5, HERPUD1, XBP1, CBS, ZNF419, KLHL24, TRIB3, ZFP69B, ATP6V1G2, VEGFA, GDF15, TUBE1, A RRDC3, CEBPG, SNORA16A, RGS4, BLOC1S5-TXNDC5, LOC390705, KIM-1, IL6, CXCL2, RELA, HSD17B11, AGPAT3, SETD1B, FTL, MAFG, IL33, SLC40A1, TF, TFRC,FTH1, HAMP, STEAP3, DRD4, DRD5, MAP3K5, SLC2A1, SLC2A3, SLC2A6, SLC2A8, SLC2A12, GLUT3, SLC2A14, EIF2AK4, EIF2S1, ATF4, ALOX5, ACSF2, IREB2, GPX4, NFE2L2, ELAVL1, SLC3 A2, TFAP2C, SP1, HBA1, NNMT, PLIN4, HIC1, STMN1, RRM2, CAPG, HNF4A, NGB, YWHAE, GABPB1, AURKA, MIR4715, RIPK1, PRDX1, MIR30B, SLC7A11, AKR1C1, AKR1C2, AKR1C3, RB1, HSPB1, HSF1, GCLC, SQSTM1, NQO1, MUC1, MT1G, CISD1, FANCD2, FTMT, HELLS, FADS2, SRC, STAT3, PML, MTOR, NFS1, TP63, CDKN1A, MIR137, ENPP2, FH, CISD2, MIR9-1, MIR9-2, MIR9-3, ISCU, ACSL3, OTUB1, CD44, LINC00336, BRD4, PRDX6, MIR17, SCD, NF2, ARNTL, JUN, CA9, TMBIM4, HSPA5, PLIN2, MIR212, Fer1HCH, AIFM2, LAMP2, ZFP36, PROM2, CHMP5, CHMP6, CAV1 and GCH1. ,

[0019] The present invention also provides a method for constructing a model for predicting the probability of suffering from Alzheimer's disease, comprising the following steps:

[0020] (8.1) Use the ssGSEA package to analyze the immune cell composition of normal samples and Alzheimer's disease samples in the training set, screen the immune-related genes that are differentially expressed in normal samples and Alzheimer's disease samples, and use the spearman algorithm to calculate the correlation between the differentially expressed immune-related genes and the correlation between the genes related to Alzheimer's disease and the immune-related gene set;

[0021] (8.2) Screening for immune gene sets related to Alzheimer's disease with a correlation coefficient > 0.3, a significant P value < 0.05, and differentially expressed immune genes, and using the immune gene sets to construct a nomogram diagnostic model, i.e., a model for predicting the probability of developing Alzheimer's disease;

[0022] The training set described in step (8.1) is the data obtained by merging the GSE5281 and GSE48350 datasets as the training set;

[0023] The gene related to Alzheimer's disease described in step (8.2) is a gene screened according to the method described above.

[0024] The present invention also provides a method for constructing a drug model that can predict the drug treatment of Alzheimer's disease, comprising the following steps:

[0025] Genes related to Alzheimer's disease were entered into the CTD database to obtain gene and drug pairs related to Alzheimer's disease; Cytoscape was used to visualize gene and drug pairs related to Alzheimer's disease, and a drug model that can predict the treatment of Alzheimer's disease was constructed.

[0026] Preferably, the gene associated with Alzheimer's disease is a gene screened according to the method described.

[0027] The present invention provides an application of GOT1, MT3, SETD1B, FH and HRAS genes and a method for screening genes related to Alzheimer's disease. The present invention screens and obtains 5 genes related to both N6-methyl and ferroptosis by the following method, namely GOT1, MT3, SETD1B, FH and HRAS genes. The screening method is: (1) merging the GSE5281 and GSE48350 data sets including normal samples and Alzheimer's disease samples as training sets; (2) extracting genes related to N6-methyl in the training set, using wilcox.test to test the obtained N6-methyl related genes, and using false positives The discovery rate (FDR) tests the pValue, screens the genes with a threshold value of fdr<0.05, and obtains the N6-methyl-related gene DE-m6A that is differentially expressed in normal samples and Alzheimer's disease samples in the training set; (3) Using the spearman algorithm, analyze the correlation between the DE-m6A obtained in step (2) and the ferroptosis-related genes, select the ferroptosis-related genes with a correlation>0.4 and a pvalue value<0.05, and obtain the gene DE-m6A-FRG that is both related to N6-methyl and ferroptosis; (4) Use the SVM-RFE and Boruta algorithms to screen the characteristic genes in DE-m6A-FRG, take the characteristic genes with intersections under the two algorithms, and obtain genes related to Alzheimer's disease. The 5 genes screened by the present invention can be used to construct a model for predicting the probability of suffering from Alzheimer's disease, and can construct a drug model for predicting the treatment of Alzheimer's disease, which provides an important reference for the diagnosis, mechanism research and treatment of Alzheimer's disease in the later stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flowchart of the analytical method is shown;

[0029] Figure 2 The PCA analysis results of the two data sets before and after batch removal (the left figure is the PCA analysis result before batch removal, and the right figure is the PCA analysis result after removing the batch effect and standardizing the data);

[0030] Figure 3 Box plots of the expression distribution of each sample in the two datasets before removing the batch effect (the first 23 from left to right are samples of the dataset GSE5281, and the last 62 are samples of the dataset GSE48350);

[0031] Figure 4 Box plots of the expression distribution of each sample in the two datasets after removing the batch effect (the first 23 from left to right are samples of the dataset GSE5281, and the last 62 are samples of the dataset GSE48350);

[0032] Figure 5 Violin plot of the expression of DE-m6A in the AD group and the control group in the training set (the fdr value represents the corrected P value, and fdr<0.05 indicates that the gene expression is significantly different between the two groups);

[0033] Figure 6 It is a circle diagram of DE-m6A on chromosome position (the red font is the DE-m6A screened by the rank sum test, and the genes marked in black font are other m6A genes with no significant difference);

[0034] Figure 7 This is the consistent clustering result diagram;

[0035] Figure 8 This is a clustering effect diagram of 2 subtype samples;

[0036] Fig. 9 Results of PCA analysis for 2 subgroups;

[0037] Fig.10 The volcano plot is for differentially expressed genes;

[0038] Fig.11 It is a heat map of differentially expressed genes among cluster subgroups (the horizontal axis represents sample information, the vertical axis represents differentially expressed genes, red represents high relative expression, and blue represents low relative expression);

[0039] Fig.12GO_KEGG enrichment bubble map (the upper picture is the GO enrichment bubble map, and the lower picture is the KEGG enrichment bubble map; the vertical axis represents the enriched GO Term and KEGG Pathway; the horizontal axis represents GeneRatio, which refers to the ratio of the number of genes related to the Term in the differential genes to the total number of differential genes; the color represents the Pvalue, the redder the color, the smaller the Pvalue; the size of the circle represents the number of genes contained in the term, and the larger the circle, the more genes it contains);

[0040] Fig.13 Heat map of DE-m6A expression in cluster subgroups;

[0041] Fig.14 Violin plot of DE-m6A expression in cluster subgroups;

[0042] Fig.15 It is a heat map of the correlation coefficient between DE-m6A and some ferroptosis-related genes (the ordinate represents 2 DE-m6A; the abscissa represents some ferroptosis-related genes);

[0043] Fig.16 The relationship between the prediction accuracy of SVM-RFE and the number of features (the horizontal axis represents the number of feature genes, the vertical axis represents the generalization error under 10-fold cross validation, and the trend of the line graph represents the relationship between the number of feature genes and the generalization error. When the line is at the lowest point, the error rate of the model is the lowest);

[0044] Fig.17 It is a box plot of the importance distribution of each gene in the Boruta algorithm (the horizontal axis represents the gene, and the vertical axis is the importance value corresponding to the gene; the blue box represents the minimum, medium and maximum importance values ​​of the shaded attribute; red represents the importance value of the rejected attribute, green represents the importance value of the confirmed attribute, and yellow represents the importance of the tentative attribute);

[0045] Fig.18 Venn diagram of the intersection of the feature genes selected by SVM-RFE and Boruta machine learning;

[0046] Fig.19 Box plot of the expression of key genes in the control group and AD group of the training set (**p<0.01; ***p<0.0001; ****p<0.00001);

[0047] Fig. 20 This is the GO enrichment result of the key gene GOT1;

[0048] Fig.21 This is the KEGG enrichment result of the key gene GOT1;

[0049] Fig. 22 The results of key gene-related disease and function enrichment (the larger the absolute value of Z-score, the darker the color of the corresponding bar graph; orange represents activated pathways (Zscore>0), the darker the color, the larger the Zscore value; blue represents inhibited pathways (Zscore<0), the darker the color, the smaller the Zscore);

[0050] Fig.23 This is the activation and inhibition pathway diagram of the key gene HRAS (the upper figure is the activation pathway, and the lower figure is the inhibition pathway);

[0051] Fig.24 The ROC curve of the LR model in the training set and the validation set (the left picture is the training set, and the right picture is the validation set);

[0052] Fig.25 The nomogram model of key genes in the training set and validation set (the left picture is the training set result, and the right picture is the validation set result);

[0053] Fig.26 Calibration curves for evaluating the predictive ability of the nomogram model in the training set and the validation set (the left figure is the calibration curve corresponding to the training set, and the right figure is the calibration curve corresponding to the validation set; idel represents the actual probability of the nomogram prediction under ideal conditions, and Bias-corrected represents the true prediction of the nomogram);

[0054] Fig. 27 Box plot of the abundance of 29 immune gene sets in the AD sample group and the Control group (the vertical axis represents the gene expression level, the horizontal axis represents the immune cell gene set, *p<0.05, **p<0.01, ***p<0.0001);

[0055] Fig.28 It is a heat map of the relationship between immune-related gene sets (the numbers represent the correlation coefficient, the redder the color, the more positive the correlation, and the bluer the color, the more negative the correlation);

[0056] Fig.29 This is a bubble chart of the correlation between AD key genes and 29 immune-related gene sets (the more orange the color, the more positive the correlation, the more purple the color, the more negative the correlation; the size of the circle indicates significance, the larger the circle, the more significant; *p<0.05, **p<0.01);

[0057] Fig.30 The nomogram model for the differential immune gene set;

[0058] Fig.31 is the ROC curve of the nomogram model in the training set;

[0059] Fig.32Calibration curve for evaluating the predictive ability of the nomogram model (idel represents the actual probability of the nomogram prediction under ideal conditions, and bias-corrected represents the true prediction of the nomogram);

[0060] Fig.33 The Venn diagram of the intersection of miRNAs predicted by the miRwalk database and the ENCORI database;

[0061] Fig.34 is the miRNA-mRNA network diagram (the red circles represent mRNA, and the blue triangles represent miRNA);

[0062] Fig.35 is the TF-mRNA network diagram (red circles represent key genes, and blue diamonds represent TFs);

[0063] Fig.36 is the TF / miRNA-mRNA network diagram (red circles represent mRNA, blue triangles represent miRNA, and yellow diamonds represent TF);

[0064] Fig.37 This is the interaction network diagram of key genes and drugs (red circles represent key genes; blue diamonds represent predicted drugs). DETAILED DESCRIPTION

[0065] The solutions provided by the present invention are described in detail below in conjunction with the embodiments, but they should not be understood as limiting the protection scope of the present invention.

[0066] Example 1

[0067] Data Source

[0068] The datasets GSE5281 and GSE48350 in the GEO database (https: / / www.ncbi.nlm.nih.gov / geo / ) are both brain tissue data, and the sequencing platforms are GPL570[HG-U133_Plus_2]AffymetrixHuman GenomeU133 Plus 2.0Array. GSE5281 and GSE48350 both include samples of brain tissues from different regions. The hippocampal tissue samples in the two datasets were extracted respectively. GSE5281 includes 13 normal samples and 10 Alzheimer's disease samples (AD samples), and GSE48350 includes 43 normal samples and 19 Alzheimer's disease samples (AD samples). The hippocampal samples of the two datasets were merged to remove the batch effect and used as the training set. The brain tissue dataset GSE33000 was downloaded from the GEO database, including 157 normal samples and 310 AD samples, as the validation set.

[0069] According to existing literature (PMID: 31100245, PMID: 29789545 and PMID: 35164788), 27 m6A methylation regulation-related genes were downloaded from the NCBI website (https: / / www.ncbi.nlm.nih.gov / ). These genes and their accession numbers in NCBI are shown in Table 1.

[0070] Table 1. m6A methylation-regulated genes and their accession numbers

[0071] name Accession Number name Accession Number name Accession Number METTL4 64863 ZCCHC4 29063 PRRC2A 7916 KIAA1429 25962 YTHDC1 91746 FMRP 2332 RBM15 64783 YTHDF1 54915 YTHDC2 64848 METTL5 29081 YTHDF2 51441 HNRNPG 27316 METTL14 57721 YTHDF3 253943 HNRNPC 3183 WTAP 9589 IGF2BP1 10642 HNRNPA2B1 3181 METTL3 56339 IGF2BP2 10644 ALKBH5 54890 METTL16 79066 IGF2BP3 10643 FTO 79068 ZC3H13 23091 eIF3 8661 RBM15B 29890

[0072] According to the records in the existing literature (PMID: 33727924), 259 ferroptosis-related genes were downloaded from the NCBI website. The specific genes and their accession numbers in NCBI are shown in Table 2.

[0073] Table 2 259 ferroptosis-related genes and accession numbers

[0074]

[0075]

[0076]

[0077] PMID: 31100245 The literature is: Shi H, Wei J, He C. Where, When, and How: Context-Dependent Functions of RNAMethylation Writers, Readers, and Erasers. Mol Cell. 2019; 74(4): 640-650.

[0078] PMID:29789545 The literature is: Yang Y, Hsu PJ, Chen YS, Yang YG.Dynamictranscriptomic m 6 Adecoration:writers,erasers,readers and functions inRNAmetabolism.CellRes.2018;28(6):616-624.

[0079] PMID: 35164788 The literature is: Deng LJ, Deng WQ, Fan SR, et al. m6A modification: recent advances, anticancertargeted drug discovery and beyond. Mol Cancer. 2022; 21(1): 52. Published2022 Feb 14.

[0080] PMID:33727924 The document is: Jiang

[0081] Example 2

[0082] Analysis methods and processes

[0083] 1. The datasets GSE5281 and GSE48350 were merged, and PCA was used to analyze the distribution of the merged data. ComBat was then used to perform batch correction on the merged data. PCA was used to perform PCA analysis on the corrected data, and a box plot was used to show the gene expression distribution of each sample before and after batch correction. The merged data was used as the training set.

[0084] 2. In the training set, the expression matrix of m6A-related genes was extracted, and the m6A-related genes obtained were tested using wilcox.test, and the pValue was tested using the false discovery rate (FDR). The differential genes were screened using the obtained fdr value, and the N6-methyl-related genes differentially expressed in normal samples and Alzheimer's disease samples in the training set were obtained, named DE-m6A, and the screening threshold was fdr<0.05. The results were displayed using heat maps and violin plots. "Omiccircos" was used to visualize the location of DE-m6A on chromosomes.

[0085] 3. Based on the obtained DE-m6A, the DE-m6A expression matrix was obtained, and the R package "ConsensusClusterPlus" was used to perform consistency clustering analysis on the AD samples in the training set. The clustering parameters maxK = 6, clusterAlg = "pam", distance = "pearson", and the best clustering method were selected. PCA analysis was performed based on the cluster subgroups obtained by clustering to check the distribution of different subgroups.

[0086] 4. Obtain all gene expression matrices in AD samples in the training set. Based on the above-obtained consistent clustering results as grouping information, use the R package “limma” to screen out differentially expressed genes (DEGs) between consistent clustering subtypes. The screening threshold is: |log2FC|>0.5, Pvalue<0.05. The results are displayed in heat maps and volcano maps.

[0087] 5. GO and KEGG enrichment analysis of DEGs was performed using the clusterProfiler package, and the enrichment results were visualized using the R package “ggplot2”.

[0088] 6. Based on the obtained DE-m6A, extract their expression in different cluster subgroups, analyze the expression level of DE-m6A in different subtypes according to the clustering results, and display the results in heat maps and violin plots. Steps 2 to 6 are used to verify that the screened DE-m6A is related to Alzheimer's disease and is related to different subgroups of Alzheimer's disease.

[0089] 7. The “spearman” algorithm was used to analyze the correlation between DE-m6A and 259 ferroptosis-related genes, and ferroptosis-related genes with correlation > 0.4 and Pvalue < 0.05 were selected to obtain genes that were both related to N6-methyl and ferroptosis and named DE-m6A-FRG.

[0090] 8. The R packages "SVM-RFE" and "Boruta" algorithms were used to screen the characteristic genes in DE-m6A-FRG respectively, and then the characteristic genes obtained under the two algorithms were intersected to obtain genes related to Alzheimer's disease, which were named AD key genes. The expression of AD key genes in normal samples and AD samples was checked in the training set, and the expression levels of AD key genes in the two groups were verified using (wilcox.test), and the results were displayed as box plots.

[0091] 9. Use the “clusterProfiler” package and the “org.Hs.eg.db” package to perform GSEA analysis of AD key genes to find the regulatory pathways and related molecular functions enriched in genes related to each key gene.

[0092] 10. Use wilcox.test to test all genes in normal samples and AD samples in the training set to obtain Pvalue, and use false discovery rate (FDR) to test Pvalue to obtain FDR value, and use FDR value to screen differential genes, with screening threshold FDR<0.05. Use IPA to analyze the classic disease pathways of differential genes, where Z-score>0 indicates that the pathway is activated, and z-score<0 indicates that the pathway is inhibited. Display the results in a bar graph; and screen all pathways containing key AD genes for display.

[0093] 11. Use the logistic regression algorithm in the training set and validation set to construct a logistic regression model, draw the ROC curve of the corresponding logistic regression model, calculate the AUC area under the ROC curve, and construct a nomogram model in the two data sets, and draw the calibration curve corresponding to the nomogram.

[0094] 12. In the training set, the ssGSEA package was used to perform immune cell infiltration analysis to observe the composition of immune cells in the training set, as well as the differentially expressed immune cells in the AD sample group and the normal sample group. The results were displayed as box plots. The "spearman" algorithm was used to calculate the correlation between immune-related gene sets and the correlation between AD key genes and immune-related gene sets, and the results were displayed using heat maps and bubble charts, respectively.

[0095] 13. Based on the above analysis, the immune gene set with a correlation coefficient with AD key genes>0.3, a significant Pvalue<0.05 and differences between the AD sample group and the normal sample group was screened out, and a nomogram diagnostic model of the immune gene set was constructed, that is, a model for predicting the probability of suffering from Alzheimer's disease; and the calibration curve corresponding to the nomogram and the ROC curve of the nomogram were drawn, and the AUC area corresponding to the ROC curve was calculated.

[0096] 14. Based on the obtained AD key genes, the miRNAs corresponding to the key genes were found in the miRWalk database (http: / / mirwalk.umm.uni-heidelberg.de / search_genes / ) and the ENCORI database (https: / / starbase.sysu.edu.cn / agoClipRNA.php?source=mRNA), and then the intersection was taken to obtain the miRNAs of the AD key genes, and the miRNA-mRNA network diagram was drawn using the Cytoscape software; the TFs corresponding to the key genes were searched using NetworkAnalyst (https: / / www.networkanalyst.ca / NetworkAnalyst / uploads / ListUploadView.xhtml), the interaction information between the key genes and TFs was obtained, and then the Cytoscape software was used for visualization; based on the obtained miRNA-mRNA relationship pairs and TF-mRNA relationship pairs, the TF / miRNA-mRNA network was constructed, and the Cytoscape software was used for visualization.

[0097] 15.In the CTD database (http: / / ctdbase.org / ), enter the AD key genes to predict potential drugs for the treatment of Alzheimer's disease, analyze the interaction between the predicted drugs and the AD key genes, use Cytoscape for visualization, and construct a drug model that can predict the treatment of Alzheimer's disease.

[0098] The specific process is as follows Figure 1 shown.

[0099] Example 3

[0100] 3.1 Dataset Merger

[0101] Extract Alzheimer's disease samples (AD samples) and normal samples (control group) in hippocampal tissue from the datasets GSE5281 and GSE48350. The dataset GSE5281 contains 13 normal samples and 10 AD samples, and GSE48350 contains 43 normal samples and 19 AD samples. Then take the intersection genes in the two datasets, a total of 22,880 genes, merge them by gene, and obtain the training set, which includes 56 normal group samples and 29 AD samples. First, perform PCA analysis on the merged data to see if there is a batch effect in the two datasets. The results are as follows: Figure 2 As shown in the figure on the left, the two data sets were clearly separated and a box plot of the expression levels of each sample was drawn, and it was clearly found that there were differences between the two data sets, such as Figure 3As shown in the figure, the ComBat method in the R software package "SVA" was used to remove the batch effect of the two data sets, and the normalizeBetweenArrays method in the "limma" package was used to standardize the data after removing the batch effect. Then, the obtained data was analyzed using the analysis method before batch removal, and PCA analysis was performed. The results are shown in the figure. Figure 2 As shown in the figure (right), the two data sets are evenly distributed and discrete, and there is no batch effect. A box plot of the standardized data expression matrix is ​​drawn, as shown in Figure 4 As shown, the results also showed that the median expression levels were on a straight line, so the processed data were used as the training set for subsequent analysis.

[0102] 3.2 Identification of DE-m6A genes

[0103] In the training set, 27 m6A-related gene expression matrices were extracted, of which only 23 genes were detected in the training set. Genes eIF3, FMRP, HNRNPG, and METTL16 were not detected in the training set. Based on the grouping information of the AD sample group and the control group, the 23 m6A-related gene expression matrices were tested using wilcox.test, and the significance value Pvalue of the difference between the two groups of samples was obtained. The false discovery rate (F DR) was used to test the Pvalue, and the differential genes were screened using the obtained fdr value. The screening threshold was fdr<0.05. A total of two differentially expressed m6As were obtained, named DE-m6As, namely WTAP and METTL14. The results were displayed in violin plots, as shown in the figure. Figure 5 Based on the location information of each gene on the chromosome, the R software package "Omiccircos" was used to draw the circos diagram of DE-m6A on the chromosome, visually displaying the location information of DE-m6A on the chromosome. The results are shown in Figure 6 shown.

[0104] 3.3 Consistency cluster analysis

[0105] Consistency cluster analysis is a common method for cancer subtype classification research. It can classify samples into several subtypes based on different omics data sets, thereby discovering new disease subtypes or performing comparative analysis of different subtypes. The idea of ​​consistency clustering is to use a resampling method to extract a certain sample data set, specify the number of clusters K, and calculate the rationality under different numbers of clusters.

[0106] Based on the training set data obtained after merging, the DE-m6A expression matrix in the AD sample group was obtained, and then the R package "ConsensusClusterPlus" was used to perform a consistent clustering analysis on the AD sample group, combined with the CDF (cumulative distribution function) cumulative distribution function (CDF describes the probability that a random variable takes a value less than or equal to a given value). The more stable the previous point of the maximum change in the CDF value and the CDF downward trend, the better the clustering effect. The best clustering method is selected, such as Figure 7 and 8 As shown in the figure, the results show that clustering into two subtypes (K=2) is the most appropriate. In order to further confirm whether there are differences between the two subgroups, PCA analysis was then performed on the two subgroups. The results showed that the classification results of the two subgroups were good, as shown in Figure 2. Fig. 9 shown.

[0107] 3.4 Analysis of differentially expressed genes among different subtypes

[0108] Differential gene analysis aims to find genes that differ between different sample groups and conduct further functional mining of differential genes. In the training set, the gene expression matrix corresponding to the AD sample group was extracted, and the consistent clustering subtypes were used as grouping information. The "limma" package was used to perform differential gene analysis between subtypes, and the screening conditions were |log2FC|>0.5, Pvalue<0.05. There were 387 significantly differentially expressed genes in cluster2-vs-cluster1, including 193 upregulated genes and 194 downregulated genes. In order to understand the distribution of differentially expressed genes as a whole, a volcano plot was used to display the distribution of differentially expressed genes. Fig.10 shown.

[0109] Fig.10 In the figure, the horizontal axis represents the difference fold (cluster2 / cluster1, taking the logarithm of 2), and the vertical axis represents -log10 (Pvalue); each point in the figure represents a gene, and the blue and red points represent significantly differentially expressed genes. The red point indicates that its gene expression is upregulated (cluster2 relative to cluster1), the blue point indicates that its gene expression is downregulated (cluster2 relative to cluster1), and the gray point indicates that these genes have no significant difference.

[0110] Heatmap is an intuitive visualization method for analyzing the distribution of experimental data. It can be used to control the quality of experimental data and intuitively display the difference in expression data of key research objects. It can also cluster data and samples to observe sample quality. It has many forms, but the basic elements are universal. Fig.11 shown.

[0111] 3.5 Enrichment analysis of differentially expressed genes in different subtypes

[0112] Gene Ontology, or GO for short, is an internationally standardized gene function classification system that provides a set of dynamically updated controlled vocabularies to comprehensively describe the properties of genes and gene products in organisms. GO has three ontologies, which describe the molecular function (MF), cellular component (CC), and biological process (BP) of genes. The basic unit of GO is Term (entry, node), and each Term corresponds to an attribute.

[0113] In a specific biological function, genes usually play a certain role by interacting with each other. Pathway analysis can help to gain a deeper understanding of the biological function of genes. KEGG is the main public database related to pathways. Using KEGG pathways as units, hypergeometric tests are applied to find metabolic pathways or signal transduction pathways that are significantly enriched in differentially expressed genes compared with the entire genome as a background, and analyze and identify them.

[0114] The “clusterProfiler” package in R was used to perform (GO and KEGG) functional enrichment analysis on DE-m6A. According to the significance threshold pvalue<0.05, 211 GO BPs, 7 GO CCs, 50 GO MFs and 30 KEGG pathways were screened out. The R package “ggplot2” was used to visualize the enrichment results. Here we select the TOP10 Descriptions of GO and KEGG for display according to the ranking of Pvalues. Fig.12 shown.

[0115] 3.6 Expression of DE-m6A in different subtypes

[0116] Based on the two DE-m6A obtained: WTAP and METTL14, the expression matrix in the two subgroups obtained by clustering was extracted, and the R software packages "pheatmap" and "ggplot2" were used to draw the heat map and violin plot of DE-6m6A expression in the two cluster subgroups. The results showed that METTL14 achieved significant differences in the two subgroups, as shown in Figure 2. Fig.13 and 14 shown.

[0117] 3.7 Identification of ferroptosis genes associated with DE-m6A

[0118] The expression matrix of 259 ferroptosis-related genes in the training set was extracted, and a total of 240 ferroptosis-related genes were detected in the training set. Based on the expression matrix obtained, the "Spearman" algorithm was used to calculate the correlation coefficient and significant Pvalue between DE-m6A and ferroptosis-related genes, and some genes with high correlation were selected for heat map display. The results are shown in Figure 2. Fig.15 As shown, with the absolute value of the correlation coefficient >0.4 and Pvalue<0.05 as the screening conditions, 16 ferroptosis genes highly correlated with DE-m6A were obtained and named DE-m6A-FRG, namely NFS1, GOT1, FH, HRAS, YWHAE, ATP6V1G2, DRD5, PEBP1, MT3, CEBPG, ATF3, TAZ, SETD1B, KLHL24, SIRT1, and ATM.

[0119] 3.8 Machine learning to screen key genes

[0120] In order to obtain the key AD genes, two machine learning algorithms were used for analysis, namely SVM-RFE and Boruta. The intersection of the characteristic genes obtained by the analysis was taken as the screened key AD genes.

[0121] 3.8.1SVM-RFE screening of characteristic genes

[0122] Support Vector Machine Recursive Feature Elimination (SVM-RFE) is an effective feature selection technique. This method is based on the sequential backward selection algorithm of the maximum margin principle of SVM. The model training samples are then sorted by the score of each feature, the feature with the smallest feature score is removed, and the model is trained again with the remaining features for the next iteration, and finally the required number of features is selected. When exploring machine learning options, it is often useful to estimate the generalization error and use it as a benchmark. When feature selection is performed on a dataset with many features, it will select some really useful features for generalization, but it may also select some useless features that are accidentally consistent with the label.

[0123] The R software package "e1071" was used to perform SVM-RFE analysis on the 16 DE-m6A-FRGs obtained above. The parameter settings were: k = 10, halve.above = 100, nfold = 10. Through SVM-RFE analysis, the relationship between SVM-RFE prediction accuracy and feature number was obtained, as shown in Fig.16 As shown in the figure, the red font part represents the number of genes when the generalization error is the lowest. Finally, a total of 10 genes were screened, including GOT1, MT3, NFS1, SETD1B, CEBPG, PEBP1, FH, HRAS, TAZ, and ATM.

[0124] 3.8.2 Boruta screening characteristic genes

[0125] Boruta is a feature selection algorithm that follows all relevant feature selection methods and can screen out all feature sets that are relevant to the dependent variable. The stability and scalability of the experimental results are very good. The screening process is as follows: First, randomness is added to the given data set by creating mixed copies of all features (i.e., shadow features). Then, it trains an extended data set for random forest classification and uses a feature importance measure (the default setting is average reduced precision) to evaluate the importance of each feature, the higher the more important. In each iteration, it checks whether a real feature has a higher importance than the best shadow feature (i.e., whether the feature has a higher score than the largest shadow feature) and continuously removes features that it considers to be very unimportant. Finally, the algorithm stops when all features are confirmed or rejected, or when the algorithm reaches a specified limit for the random forest operation. This method minimizes the error of the random forest model, which will eventually form a minimized optimal feature subset. On the other hand, Boruta finds all features, regardless of their strong or weak correlation with the decision variable, which makes it very suitable for application in the biomedical field. Some people are interested in understanding which human genes (features) are associated with specific medical conditions (target variables) to some extent.

[0126] The R software “Boruta” package was used to perform Boruta analysis on the 16 DE-m6A-FRGs obtained above. The parameter settings were: pValue=0.05, mcAdj=T, maxRuns=300, and characteristic genes were screened. The results are as follows Fig.17 As shown, green indicates the importance value of the confirmed attribute, which is the characteristic gene obtained by screening. Finally, 8 characteristic genes were screened: GOT1, FH, HRAS, ATP6V1G2, MT3, SETD1B, KLHL24, and SIRT1.

[0127] 3.8.3 Identification of key genes

[0128] Using the online jveen graph production website (http: / / jvenn.toulouse.inra.fr / app / example.html), we took the intersection of the characteristic genes screened by SVM-RFE and Boruta machine learning, and obtained a total of 5 overlapping genes, namely GOT1, MT3, SETD1B, FH, and HRAS, which are the key AD genes screened. The results are shown in the figure. Fig.18As shown. In the training set, the expression of key genes in the control group and AD sample group was checked, and the expression of key genes in the two groups was verified using wilcox.test. It was found that the expression of 5 genes between the control group and the AD sample group was significantly different. The R software package "ggplot2" was used for visualization, and the results were displayed in box plots, as shown in the figure below. Fig.19 shown.

[0129] 3.9 Functional enrichment of key genes

[0130] In the training set, the Pearson correlation coefficient between each key gene and all genes in the data set was calculated, and then the correlation coefficients were sorted from large to small. After each key gene obtained the corresponding gene ranking list, the R package "clusterProfiler" was used to perform GO and KEGG GSEA enrichment analysis. The Benjamini&Hochberg method was used for multiple test correction to obtain the corrected p value, i.e., p.adjust. p.adjust<0.05 was considered a significant enrichment result. According to the significance ranking, the GO and KEGG TOP5 Descriptions were selected for visualization. Fig. 20 and Fig.21 The enrichment results of the key gene GOT1 are shown respectively.

[0131] 3.10IPA analysis

[0132] IPA (IPA Pathway Analysis) is an integrated analysis software for analyzing gene expression data, microRNA data and small-scale experimental data. A visual experimental system was established to construct a network of interactions between molecules based on the properties of various molecules such as genes, proteins, chemicals, and drugs. IPA can provide multiple analysis contents such as classic signal pathway analysis, upstream regulation analysis, disease and function analysis, regulatory effect analysis, and gene interaction network analysis. To identify typical pathways, diseases, and functional pathways related to key genes, firstly, all genes in the training set were subjected to a wilcox.test test between the control group and the AD sample group to obtain the significant value Pvalue of the difference between the two groups of samples for each gene, and the false discovery rate (FDR) was used to test the Pvalue, and the differentially expressed genes were screened using the obtained fdr value. The screening threshold was fdr<0.05, and a total of 2481 differentially expressed genes were obtained. IPA functional enrichment analysis was performed based on the differentially expressed genes obtained. Z-score is a unique algorithm developed by IPA to predict the activation or inhibition status of a pathway. Generally speaking, Z-score>0 indicates that the pathway is activated, and Z-score<0 indicates that the pathway is inhibited. In this study, there are 76 inhibited pathways and 68 activated pathways. They are sorted from large to small according to enrichment significance -log10 (Pvalue), and a bar graph of each pathway is drawn. Fig. 22 The top 10 pathways are shown, with blue bars representing inhibited pathways and orange representing activated pathways. GOT1 is involved in 1 inhibited pathway: Aspartate Degradation II (aspartate degradation, Z-score = -2.000); FH is involved in 1 inhibited pathway: TCA Cycle II (Eukaryotic) (TCA cycle, Z-score = -3.317); HRAS is involved in 37 activated pathways and 29 inhibited pathways, of which Pulmonary Healing Signaling Pathway (pulmonary healing signaling pathway, Z-score = 3.651) and Synaptogenesis Signaling Pathway (synaptogenesis signaling pathway, Z-score = -3.833) are the two pathways with the strongest activation and inhibition strengths involved by gene HRAS, such as Fig.23 shown.

[0133] 3.11Construction of AD key gene diagnostic model

[0134] 3.11.1 Construction of key gene LR model

[0135] Logistic Regression (LR) is based on the linear regression model and uses the sigmoid function to compress the result of the linear model wTx to between [0,1] to make it have probabilistic meaning. LR is an algorithm proposed to solve the binary classification problem. Its essence is still a linear model and its implementation is relatively simple. It is frequently used in advertising calculations and recommendation systems and is the basic algorithm of the CTR estimation model. At the same time, the LR model is also a basic component of deep learning.

[0136] The receiver operating characteristic curve (ROC curve) is also called the sensitivity curve. There is no correlation between the horizontal and vertical coordinates of the ROC curve, so the ROC curve cannot be analyzed as a function curve. The ROC curve should be regarded as countless points, each of which represents a classifier, and its horizontal and vertical coordinates represent the performance of this classifier. AUC (Area Under Curve) is usually used as an indicator for classifier performance evaluation, where AUC refers to the area under the ROC curve, and the value range is generally between 0.5 and 1. The larger the value, the better the performance of the classifier.

[0137] This study combined the LR model with the ROC curve to study the diagnostic performance of 5 key genes for disease or not. In the training set, the LR model was first constructed for the 5 key genes using the R software package "glmnet", and the parameter family was set to "binomial". Then the ROC curve of the LR model was drawn using "pROC". The area under the ROC curve was used to determine the diagnostic performance of the LR model. The results showed that the AUC size in the training set was 0.873, as shown in Figure 2. Fig.24 As shown in the left figure, the same method was used in the validation set GSE33000 to calculate the area under the ROC curve of the LR model of the five key genes in the validation set. The corresponding AUC was as high as 0.904. Fig.24 As shown in the figure on the right, the LR model constructed with 5 key genes has high classification performance in both the training set and the validation set.

[0138] 3.11.2 Construction of key gene nomogram

[0139] Based on the five key genes obtained from the above analysis, the nomogram model of AD key genes was constructed using the “RMS” package in the training set and validation set GSE33000, combined with the sample grouping information and the expression levels of key genes, as shown in the following figure: Fig.25 shown.

[0140] In order to verify the accuracy of the nomogram model in predicting AD, based on the nomogram constructed above, the calibration curves in the training set and the validation set GSE33000 were drawn using the calibrate function and the boot method. The results are shown in Figure 2. Fig.26 As shown, the closer the slope of the calibration curve is to 1, the more accurate the prediction is. The calibration curves in the training set and the validation set show that the error between the actual risk of disease and the predicted risk is very small, indicating that the nomogram model has a high prediction accuracy for AD.

[0141] 3.12 Immune Infiltration Analysis

[0142] ssGSEA (single sample GSEA) is a single sample GSEA method. Through ssGSEA analysis, the activity of immune cells, immune functions, and immune pathways of each sample can be obtained, and then grouped according to immune activity. ssGSEA is mainly an implementation method proposed for single samples that cannot perform GSEA. The principle is similar to GSEA. ssGSEA calculates the rank value of each gene based on the expression profile file, and then performs subsequent statistical analysis. A set of 29 immune-related genes (16 immune cells and 13 immune-related pathway gene sets) is selected, which includes immune cell types, immune-related pathways, and immune-related functions. This gene set can accurately obtain the immune activity of each sample.

[0143] Through ssGSEA analysis, the expression level box plots of 29 immune gene sets in each sample in the training set were drawn, and the rank sum test was performed using Wilcoxon.test. It was found that there were 9 immune-related gene sets with significant differences in expression in the training set, namely B_cells, CCR, MHC_class_I, Neutrophils, Parainflammation, pDCs, T_cell_co-inhibition, TIL, and Treg. The results are shown in Fig. 27 The “spearman” method was used to calculate the correlation of the immune-related gene set, and the corresponding Pvalue and correlation coefficient cor were obtained. The R package “ggcorrplot” was used for visualization, as shown in Fig.28 The “spearman” method was used to calculate the correlation between AD key genes and 29 immune-related gene sets, and the R software package “ggplot2” was used to draw a bubble chart for display, as shown in Fig.29As shown, it was found that the key gene HRAS was highly correlated with both CCR and pDCs, with correlation coefficients of 0.335 and 0.334, respectively, p<0.001. As can be seen from the figure, CCR and pDCs also had significant infiltration differences between the AD sample group and the control group. Therefore, there are a total of 2 immune cells that have a high correlation with the key gene and have significant infiltration differences between the AD sample group and the control group in the training set, namely CCR and pDCs.

[0144] 3.13 Construction of immune cell diagnostic model

[0145] Based on the above immune infiltration analysis, it can be seen that there are significant infiltration differences between CCR and pDCs in the training set groups, and they are highly correlated with the key AD gene HRAS (cor>0.3, Pvalue<0.001). Therefore, CCR and pDCs were selected to construct an immune cell nomogram in the training set, and the differential immune gene set was used to build a model to predict the probability of disease in the research subjects.

[0146] Combining the sample grouping information in the training set and the percentage of CCR and pDCs in each sample in the training set, the “RMS” package was used to construct a nomogram model of the differential immune gene set. The results are shown in the figure. Fig.30 As shown, the ROC curve of the nomogram model was then drawn using "pROC". The diagnostic performance of the nomogram model was determined by calculating the area under the ROC curve (AUC). The results showed that the AUC was 0.756, as shown in Fig.31 As shown, the nomogram model has a higher diagnostic performance.

[0147] In order to verify the accuracy of the nomogram model in predicting AD, a calibration curve was drawn based on the nomogram constructed above using the calibrate function and the boot method. The results are shown in Fig.32 As shown, the closer the slope of the calibration curve is to 1, the more accurate the prediction is. The calibration curve shows that the error between the actual risk of disease and the predicted risk is very small, indicating that the nomogram model has a high prediction accuracy for AD.

[0148] 3.14 TF-miRNA-mRNA network construction

[0149] 3.14.1 miRNA-mRNA prediction analysis

[0150] Based on the five key genes (GOT1, MT3, SETD1B, FH and HRAS) obtained above, miRwalk3.0 database (http: / / mirwalk.umm.uni-heidelberg.de / search_genes / ) was used to predict miRNAs for the five key genes. The default parameters were used, i.e., binding probability>=0.95; binding site position: 3UTR, and 616 mRNA-miRNA pairs were screened, including 544 miRNAs and 5 mRNAs.

[0151] Using the ENCORI database (https: / / starbase.sysu.edu.cn / agoClipRNA.php?source=mRNA), the default parameters, i.e. CLIP-Data>=1, Degradome-Data>=0, pan-Cancer>=0, were used to predict miRNAs for the five key genes, and 235 mRNA-miRNA pairs were obtained, including 196 miRNAs and 5 mRNAs. Using online jvenn, the intersection of the miRNAs obtained from the two databases was obtained, and a total of 37 miRNAs were obtained, of which 2 miRNAs corresponded to the FH gene, 15 miRNAs corresponded to GOT1, 3 miRNAs corresponded to HRAS, 2 miRNAs corresponded to MT3, and 15 miRNAs corresponded to SETD1B. The results are as follows Fig.33 As shown, the miRNA-mRNA network diagram was drawn using Cystocape software, as shown in Fig.34 shown.

[0152] 3.14.2TF Prediction Analysis

[0153] Using NetworkAnalyst (https: / / www.networkanalyst.ca / NetworkAnalyst / uploads / ListUploadView.xhtml), parameter settings: Specifyorganism = "H.sapiens (human)", Set ID type = "Official Gene Symbol", input key genes GOT1, MT3, SETD1B, FH and HRAS, retrieved 200 pairs of TF-mRNA interaction relationships, including 142 TFs and 4 mRNAs, and visualized using Cytoscape software. The results are shown in Figure 2. Fig.35 shown.

[0154] 3.14.3 TF / miRNA-mRNA network construction

[0155] Based on the 37 miRNA-mRNA intersection pairs and 200 TF-mRNA pairs obtained above, 237 TF / miRNA-mRNA pairs were obtained, including 37 miRNAs, 5 mRNAs and 142 TFs. The results were visualized using Cytoscape software. Fig.36 shown.

[0156] 3.15 Prediction of Potential Drugs

[0157] In the CTD database, key genes GOT1, MT3, SETD1B, FH and HRAS were input to predict potential drugs for the treatment of AD. A total of 71 gene-drug pairs were obtained, and then visualized using Cytoscape. The results are shown below: Fig.37 shown.

[0158] As can be seen from the above examples, the present invention provides an application of GOT1, MT3, SETD1B, FH and HRAS genes and a method for screening genes related to Alzheimer's disease. The five genes screened by the present invention can be used to construct a model for predicting the probability of suffering from Alzheimer's disease, and a drug model for predicting the treatment of Alzheimer's disease can be constructed, providing an important reference for the diagnosis, mechanism research and treatment of late Alzheimer's disease.

[0159] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. Use of one or more of GOT1, MT3, SETD1B, FH and HRAS genes in the preparation of drugs for preventing and / or treating Alzheimer's disease.

2. Application of one or more of the GOT1, MT3, SETD1B, FH and HRAS genes in constructing a model for predicting the probability of developing Alzheimer's disease.

3. Use of one or more of the GOT1, MT3, SETD1B, FH and HRAS genes in the preparation of a product that can predict drugs for treating Alzheimer's disease.

4. A method for screening genes related to Alzheimer's disease, characterized in that: The steps include: (1) The GSE5281 and GSE48350 datasets, which include normal samples and Alzheimer's disease samples, were merged as training sets; (2) Genes related to N6-methyl in the training set were extracted, and the obtained N6-methyl-related genes were tested using wilcox.test. The false discovery rate (FDR) was used to test the pValue, and genes with a threshold of fdr < 0.05 were screened to obtain the N6-methyl-related gene DE-m6A differentially expressed in normal samples and Alzheimer's disease samples in the training set; (3) Using the spearman algorithm, the correlation between the DE-m6A obtained in step (2) and ferroptosis-related genes was analyzed, and ferroptosis-related genes with correlation > 0.4 and P value < 0.05 were selected to obtain genes DE-m6A-FRG that were both related to N6-methyl and ferroptosis; (4) The SVM-RFE and Boruta algorithms were used to screen the characteristic genes in DE-m6A-FRG respectively, and the characteristic genes with intersection under the two algorithms were selected to obtain genes related to Alzheimer's disease.

5. The method according to claim 4, characterized in that The GSE5281 and GSE48350 datasets described in step (1) are both from the GEO database.

6. The method according to claim 5, characterized in that The genes related to N6-methyl in step (2) include METTL4, KIAA1429, RBM15, METTL5, METTL14, WTAP, METTL3, METTL16, ZC3H13, ZCCHC4, YTHDC1, YTHDF1, YTHDF2, YTHDF3, IGF2BP1, IGF2BP2, IGF2BP3, eIF3, PRRC2A, FMRP, YTHDC2, HNRNPG, HNRNPC, HNRNPA2B1, ALKBH5, FTO and RBM15B.

7. The method according to claim 6, characterized in that The ferroptosis-related genes in step (3) include RPL8, ATP5MC3, CS, EMC2, NOX1, CYBB, NOX3, NOX5, DUOX1, DUOX2, G6PD, PGD, VDAC2, PIK3CA, FLT3, SCP2, TP53, LPCAT3, NRAS, KRAS, HRAS, TFR2, GOT1, CARS1, ATG7, NCOA4, ALOX12B, ALOX15B, ALOXE3, PHKG2, ACO1, SLC38A1, GLS2, G6PDX, ULK1, ATG3, ATG4D, ATG5, BECN1, MAP1LC3A, GABARA PL2, GABARAPL1, ATG16L1, WIPI1, WIPI2, SNX4, ATG13, ULK2, SAT1, EGFR, MAPK3, MAPK1, BID, ZEB1, KEAP1, DPP4, ALOX15, CDKN2A, PEBP1, SOCS1, CDO1, MY B. MAPK8, MAPK9, SLC1A5, CHAC1, MAPK14, LINC00472, PRKAA2, PRKAA1, ABCC1, MIR6852, ACVR1B, TGFBR1, BAP1, EPAS1, HILPDA, HIF1A, ALOX12, ACSL4, HM OX1, IFNG, ANO6, LPIN1, HMGB1, TNFAIP3, TLR4, NOX4, ATF3, ATM, YY1AP1, EGLN2, MIOX, TAZ, MTDH, IDH1, SIRT1, FBXW7, PANX1, DNAJB6, BACH1, LONP1, PTG S2, DUSP1, NOS2, NCF2, MT3, UBC, ALB, TXNRD1, SRXN1, GPX2, BNIP3, OXSR1, SELENOS, ANGPTL7, DDIT4, LOC284561, ASNS, TSC22D3, DDIT3, JDP2, SESN2, SL C1A4, PCK2, TXNIP, VLDLR, GPT2, PSAT1, LURAP1L, SLC7A5, HERPUD1, XBP1, CBS, ZNF419, KLHL24, TRIB3, ZFP69B, ATP6V1G2, VEGFA, GDF15, TUBE1, ARRDC3 , CEBPG, SNORA16A, RGS4, BLOC1S5-TXNDC5, LOC390705, KIM-1, IL6, CXCL2, RELA, HSD17B11, AGPAT3, SETD1B, FTL, MAFG, IL33, SLC40A1, TF, TFRC, FTH1,<h2 style=";text-align:left;direction:ltr">HAMP、STEAP3、DRD4、DRD5、MAP3K5、SLC2A1、SLC2A3、SLC2A6、SLC2A8、SLC2A12、GLUT3、S LC2A14、EIF2AK4、EIF2S1、ATF4、ALOX5、ACSF2、IREB2、GPX4、NFE2L2、ELAVL1、SLC3A2、T FAP2C、SP1、HBA1、NNMT、PLIN4、HIC1、STMN1、RRM2、CAPG、HNF4A、NGB、YWHAE、GABPB1、AU RKA、MIR4715、RIPK1、PRDX1、MIR30B、SLC7A11、AKR1C1、AKR1C2、AKR1C3、RB1、HSPB1、HSF 1、GCLC、SQSTM1、NQO1、MUC1、MT1G、CISD1、FANCD2、FTMT、HELLS、FADS2、SRC、STAT3、PML 、MTOR、NFS1、TP63、CDKN1A、MIR137、ENPP2、FH、CISD2、MIR9-1、MIR9-2、MIR9-3、ISCU、AC SL3、OTUB1、CD44、LINC00336、BRD4、PRDX6、MIR17、SCD、NF2、ARNTL、JUN、CA9、TMBIM4、H SPA5、PLIN2、MIR212、Fer1HCH、AIFM2、LAMP2、ZFP36、PROM2、CHMP5、CHMP6、CAV1、GCH1。、 8. A method for constructing a model for predicting the probability of suffering from Alzheimer's disease, characterized in that: The steps include: (8.1) Use the ssGSEA package to analyze the immune cell composition of normal samples and Alzheimer's disease samples in the training set, screen the immune-related genes that are differentially expressed in normal samples and Alzheimer's disease samples, and use the spearman algorithm to calculate the correlation between the differentially expressed immune-related genes and the correlation between the genes related to Alzheimer's disease and the immune-related gene set; (8.2) Screening for immune gene sets related to Alzheimer's disease with a correlation coefficient > 0.3, a significant P value < 0.05, and differentially expressed immune genes, and using the immune gene sets to construct a nomogram diagnostic model, i.e., a model for predicting the probability of developing Alzheimer's disease; The training set described in step (8.1) is the data obtained by merging the GSE5281 and GSE48350 datasets as the training set; The gene associated with Alzheimer's disease in step (8.2) is a gene screened according to the method described in any one of claims 4 to 7.

9. A method for constructing a drug model that can predict the drug treatment of Alzheimer's disease, characterized in that: The steps include: Genes related to Alzheimer's disease were entered into the CTD database to obtain gene and drug pairs related to Alzheimer's disease; Cytoscape was used to visualize gene and drug pairs related to Alzheimer's disease, and a drug model that can predict the treatment of Alzheimer's disease was constructed.

10. The construction method according to claim 9, characterized in that: The gene associated with Alzheimer's disease is a gene screened according to the method according to any one of claims 4 to 7.

Images