The mediating role of genes in the influence of intestinal flora on type 2 diabetes and the screening of diagnostic markers Authors: Yu‐yang Chen, Yi‐fan Shi, Xin‐yue Zhang, Jing‐yue Zhang, Shi‐mao Zhang, Heng‐xia Zhao, De‐liang Liu Published: 2026 ABSTRACT ABSTRACTIntroductionThe composition of intestinal flora affects the occurrence and development of type 2 diabetes to some extent, with dysregulation of the microbiome being a clinical manifestation of the disease.MethodsThe key intestinal flora with causal relationship to type 2 diabetes was obtained by mendelian randomization (MR) analysis. Meanwhile, differentially expressed genes (DEGs) were analyzed in the GSE76894 dataset and overlapped with causal genes. The differentially expressed causal genes were screened by machine learning, and diagnostic markers of type 2 diabetes were identified. We also analyzed the expression of diagnostic markers for different cell types using the single‐cell data to provide a more reliable basis for disease diagnosis.ResultsThe study identified 10 key intestinal flora that were causally linked to type 2 diabetes through MR analysis. Transcriptome‐wide association study (TWAS) and cis‐expression quantitative trait loci.(eQTL) MR analysis of type 2 diabetes identified 228 genes that were causally linked to type 2 diabetes (|Z score| > 1, P < 0.05). About 20 causal genes were selected by MR analysis of 228 genes and intestinal flora causally related to type 2 diabetes. The diagnostic markers screened by machine learning were verified by the area under the curve (AUC), the results were all >0.7, indicating good diagnostic efficiency. Single‐cell analysis suggested that BEND7 was specifically high expression in the control group, and C1orf85 was specifically high expression in samples with type 2 diabetes.ConclusionsThe study provided a solid theoretical basis for further understanding of the underlying mechanisms of type 2 diabetes pathogenesis and progression.Keywords: Gut microbiota, Mendelian randomization, Type 2 diabetes INTRODUCTION Type 2 diabetes, a prevalent metabolic disorder characterized by glucose and insulin metabolism abnormalities 1 , has seen a global increase in prevalence, affecting hundreds of millions of people. In China, the diabetes prevalence rate has risen steadily, reaching 11.2% between 2015 and 2017 2 . Early‐stage type 2 diabetes patients may experience mild or no noticeable symptoms, primarily manifesting as thirst, fatigue, and weight loss. However, prolonged hyperglycemia can lead to serious complications, including cardiovascular diseases, diabetic nephropathy, and diabetic foot, significantly impacting quality of life and lifespan 3 . While type 2 diabetes is typically caused by a combination of genetic factors and environmental elements like unhealthy diets, physical inactivity, and obesity 4 , its exact etiology remains unclear. Treatment strategies emphasize comprehensive individualized management combining lifestyle modifications with pharmacological interventions 5 . Although these measures effectively control disease progression, they cannot cure the condition and carry risks of multiple complications. Therefore, the urgent efforts are needed to deepen research into its pathogenesis, develop more effective treatments, and promote prevention‐oriented public health strategies. In recent years, the role of the gut microbiome in Diabetes has gained more attention 6 , 7 . The human gastrointestinal tract contains a rich and complex community of microbes, including bacteria, fungi, viruses and archaea, generally known as the gut microbiome 6 . The gut microbiota is a complex microbial community residing in the gastrointestinal tract of humans and animals, participating in energy metabolism and immune regulation through symbiotic relationships 8 , 9 . Under normal conditions, Bacteroidota, Bacillota, Pseudomonadota, and Actinomycetota dominate the intestinal environment, with Bacteroidota and Bacillota maintaining relatively stable proportions. However, disruption of this balance may lead to various diseases, including non‐alcoholic steatohepatitis and type 2 diabetes 10 . Studies indicate that high‐fat diets and overnutrition compromise gut microbiota diversity and stability, reducing beneficial bacteria while increasing opportunistic pathogens—a microecological imbalance strongly associated with type 2 diabetes 11 . Additionally, Chen et al. found that in spontaneous model rats with type 2 diabetes (Goto‐Kakizaki), administration of Coptis chinensis and Rheum palmatum significantly improved glucose and lipid metabolism disorders. Gut microbiota analysis revealed that these herbs regulated microbial composition and altered relative abundance, with these changes closely linked to metabolic improvements. Furthermore, research on specific probiotics like Acinetobacter mucile demonstrates their significant impact on obesity and related diseases 12 . Meanwhile, intestinal barrier dysfunction provides a pathway for bacterial endotoxins to invade, further exacerbating insulin resistance and the progression of type 2 diabetes 13 . Taken together, these studies clearly demonstrate that gut microbiota imbalance is closely associated with the development and progression of type 2 diabetes. By regulating microbial composition and function, this approach holds promise as a potential therapeutic strategy for the management of type 2 diabetes. Therefore, this study aimed to investigate the relationship between gut microbiota and type 2 diabetes, as well as genes associated with type 2 diabetes through two‐sample Mendelian randomization (MR) analysis. Biomarkers with diagnostic value were identified through machine learning methods, followed by expression and receiver operating characteristic (ROC) analyses. Single‐cell RNA sequencing (scRNA‐seq) analysis was employed to examine expression patterns of these biomarkers across different cell types. Through this comprehensive approach, the study aimed to provide valuable references for exploring similar pathogenic mechanisms and treatment strategies for type 2 diabetes. MATERIALS AND METHODS Study design This study followed the Mendelian reporting specifications for randomized studies (STROBE‐MR) 14 . The main process was summarized as follows: the causal relationship of intestinal flora to type 2 diabetes; the causal relationship between genes and type 2 diabetes was determined by TWAS and eQTL MR analysis; the mediating of genes in the pathway from intestinal flora to type 2 diabetes. In this study, single nucleotide polymorphisms (SNPs) were used as instrumental variables (IVs), and MR analysis followed three assumptions: IVs were closely correlated with exposure factors, IVs had no causal relationship with other confounders, and IVs could not directly affect outcomes but only through exposure factors. Data source The study obtained the GWAS data of type 2 diabetes from FinnGen database (https://r10.finngen.fi/pheno/T2D) and downloaded the finngen_R10_T2D dataset, which includes 65,085 cases of type 2 diabetes and 335,112 control samples with European ancestry. Genetic variations of intestinal flora were sourced from an extensive genome‐wide meta‐analysis carried out by the MiBioGen consortium (https://mibiogen.gcc.rug.nl/menu/main/home/), which encompassed 18,340 participants from 24 groups, primarily of European ancestry (n = 13,266). The intestinal flora were grouped by the target key regions of the 16 s rRNA gene (V1‐V2 and V4, V3‐V4), and Spearman correlation analysis was used to determine gene loci affecting bacterial abundance. Finally, 211 kinds of intestinal flora traits were obtained, comprising 9 phyla, 16 classes, 20 orders, 35 families, and 131 genera. The expression quantitative trait loci (eQTL) dataset of blood was obtained from the eQTLGen database (https://www.eqtlgen.org/phase1.html), containing 16,989 genetic cis‐eQTLs derived from 31,684 blood samples of control individuals of European ancestry. Additionally, the scRNA‐seq dataset (GSE221156) of type 2 diabetes was retrieved from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), comprising islet cells from 17 controls and 14 patients with type 2 diabetes. The ENSEMBL ID was converted into Gene SYMBOL using the clusterProfiler package, and the expression profile was normalized using log2(exp + 1). GSE76894 dataset was used as the training set for differential expression analysis and GSE164416 served as the validation dataset. Detailed information of datasets used was provided in Table S1. Selection of IVs To improve the accuracy and effectiveness of causal relationship, this study implemented stringent quality control measures for the selection of IVs. (1) SNPs significantly associated with the intestinal flora were selected (P < 5.0 × 10−6). (2) The strength of each IV was evaluated using the F statistic (F = β2/SE2), and IVs showing bias trending toward zero still required filtering based on the F statistic. (3) To assure the independence of IVs and mitigate linkage disequilibrium (LD) effects, an R2 threshold of 0.01 was set for SNPs, along with a genetic distance criterion of 10 Mb. (4) For palindromic SNPs, the forward strand allele was determined based on allele frequency data. MR analysis MR analysis of intestinal flora and type 2 diabetes In order to screen the intestinal flora that had a causal relationship with type 2 diabetes, inverse variance weighting (IVW) 15 , Weighted median (WM) 16 , weighted model 17 , sample mode 18 , and MR‐Egger 18 were used to implement two‐sample MR analysis. The IVW served as the primary method, which provided an unbiased estimate when horizontal pleiotropy was not present. Only those intestinal flora for which the direction of the odds ratios (OR) was consistent across the five algorithms and where the P‐value for the IVW algorithm was less than 0.05 were considered to have a causal relationship with type 2 diabetes. The wald ratio estimates from each IVs were integrated into a meta‐analysis, with the IVs intercept constrained to zero. Statistical significance (P < 0.05) indicated that a causal relationship was possible in MR. Finally, the reverse analysis of MR was performed using the same analytical approach, excluding intestinal flora variations influenced by type 2 diabetes. MR analysis of genes and type 2 diabetes To further understand the link between genes and type 2 diabetes, a transcriptome‐wide association study (TWAS) FUSION analysis was performed on autosomes based on the Genotype‐Tissue Expression (GTEx) (http://www.gtexportal.org/home%5B33) project (V8). Based on the results of genotyping and expression levels in the reference project, the prediction model between multiple SNPs and expression levels was constructed by Elastic net, Top 1 arithmetic, Best linear unbiased prediction (BLUP), Bayesian sparse linear mixed model (BSLMM), and LASSO. The distance between the model and SNPs at each GWAS locus was estimated. FUSION was used to integrate several models and evaluate the effect of SNP on gene expression weight, and the model with the highest predictive performance was used to determine gene weight. Finally, the genetic effect score of type 2 diabetes was combined with gene weight for TWAS analysis, and results identified for the differences were screened (P < 0.05). In order to further analyze the causal relationship between type 2 diabetes and genes obtained from TWAS analysis, the cis‐eQTL data of type 2 diabetes‐related genes were used as exposure and type 2 diabetes as outcome for MR analysis. We selected all SNPs within 1 MB upstream and downstream of the gene's chromosomal location, set a genome‐wide significance threshold (FDR <0.05) to generate IVs, and then SNPs exhibiting linkage disequilibrium in each eQTL were excluded using European samples from the Thousand Genomes Project (r 2 < 0.01, kb = 10,000). The MR Estimates for each SNP were calculated using the Wald ratio method. When multiple SNPs are present, the overall estimate of the ratios is calculated as a weighted average, with weights based on the inverse variance of each ratio estimate. When the number of SNPS was no less than 3, we evaluated whether the MR‐Egger intercept deviated significantly from 0 to test for horizontal pleipotency, and heterogeneity between Wald ratios was detected by Q test. The genes selected by TWAS were intersected with the results of MR analysis, and the genes causally linked to T2D were obtained for further analysis. MR analysis of gene and intestinal flora In order to investigate the effects of genes causally linked to type 2 diabetes on intestinal flora, we took intestinal flora as exposure factors and genes causally linked to type 2 diabetes as outcomes and conducted MR analysis using cis‐eQTL data of genes and GWAS Summary data of 10 intestinal flora. The resulting genes with causal relationships with both intestinal flora and type 2 diabetes were named as causal genes. Sensitivity analysis In the sensitivity analyses, Cochran's Q test was first employed to evaluate the heterogeneity of estimates obtained from individual SNPs. Afterward, the MR‐Egger intercept test was performed in order to assess potential horizontal pleiotropy among the SNPs. A leave‐one‐out sensitivity analysis was then carried out by sequentially removing one SNP one time to evaluate whether any changes would influence the connection of exposure and outcome variables. Finally, the results are visualized and any potential horizontal pleiotropy is identified graphically. Enrichment analysis The ClusterProfiler package (v 4.7.1.001) 19 was utilized for Gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) enrichment analyses to predict the potential molecular functions of selected genes. GO enrichment included biological processes (BPs), cellular components (CCs), and molecular functions (MFs). The results of P < 0.05 were selected and visualized through the ggpubr package (v.0.5.0) 20 . Differential expression analysis The GSE76894 dataset served as training set, while GSE164416 dataset served as validation set. For the GSE76894 dataset, we annotated the data using the platform annotation file. When several probes were linked to the same gene, the expression level of that gene was determined by calculating the average value. For the GSE164416 dataset, we utilized the ClusterProfiler package to convert ENSEMBL IDs into a Gene Symbols and normalized the expression profile using log2(exp + 1). After standardizing, annotating, and cleaning the dataset, the genes expression in GSE76894 was compared in normal control and samples with type 2 diabetes by using the Wilcoxon test. The differentially expressed genes (DEGs) obtained were intersected with causal genes to identify the differential causal genes for further analysis. Immune infiltration and drug prediction analyses The single sample gene set enrichment analysis (ssGSEA) algorithm was employed to assess the relative abundance of 22 immune cell infiltrations in the microenvironment of patients with type 2 diabetes compared to normal control samples. The gene set 21 was obtained through the study of Charoentong to characterize the types of infiltrated immune cells in each tumor immune microenvironment (TIME), and this gene set was rich in various of human immune cell subtypes, such as the activated CD8 T cells, activated dendritic cells, macrophages, natural killer T cells, and regulatory T cells. The Spearman algorithm was employed in order to analyze the relationship of differential causal genes and immune cell infiltration scores. Additionally, the DGIdb database (https://www.dgidb.org/) was utilized to search for drugs that target differentially expressed causal genes. Screening diagnostic markers and constructing diagnostic models LASSO, Support vector machine recursive feature elimination (SVM‐RFE), and random forest (RF) were employed to screen for potential diagnostic markers in the training set. The ggvenn package (v0.1.9) 22 was utilized to identify the intersection genes identified by the machine learning algorithms. The ROC curve was then generated using the pROC software package, and the area under the ROC curve (AUC) was calculated to screen diagnostic markers and verify their diagnostic effectiveness. The differential expression and predictive reliability of these diagnostic markers were further validated in the GSE164416 validation set. ScRNA‐seq analysis Based on transcriptomic analysis, we conducted RNA‐seq analysis on the GSE221156 dataset to elucidate the cellular mechanisms of type. All analyses were performed by Seurat package (v 5.1.0) 23 . We applied rigorous quality control protocols to eliminate cells with fewer than 200 genes, more than 6,000 genes, or a mitochondrial gene percentage exceeding 20% or a total gene count exceeding 40,000. Using the ‘NormalizeData’ function, we normalized the RNA‐seq dataset and applied the variance stabilization transform (VST) method to identify the first 2,000 highly variable genes for the further analysis. Principal component analysis (PCA) was performed utilizing the JackStrawPlot and JackStraw functions, allowing us to select the appropriate principal components for further analysis. The t‐SNE algorithm was employed to visualize. Cell clusters were determined by K‐nearest neighbor (KNN) algorithm along with the ‘FindClusters’ function with a resolution of 0.8. Finally, cell populations were annotated based on the researches of Joshua Chiou, Nathan Lawlor, and Chun Su et al. 24 , 25 , 26 . DEGs between different cell types were identified by ‘FindAllMarkers’ function in Seurat, with parameters set to min.pct = 0.25, logfc.threshold = 0.25 while retaining only genes with P < 0.05. The FeaturePlot and VlnPlot functions were utilized to visualize the results. Statistical analysis All analyses were conducted using R software (v.4.2.2). Group differences were assessed with the Wilcoxon test, and a P‐value of <0.05 was deemed statistically significant. In the graphical representations, ns indicates P > 0.05, * indicates P < 0.05, ** indicates P < 0.01, *** indicates P < 0.001, **** indicates P < 0.0001. RESULTS A causal relationship between 10 kinds of intestinal flora and type 2 diabetes A total of 1,512 SNPs were screened as IVs for 211 microflora, with the F statistics of all IVs ranging from 17.17 to 88.43, surpassing the threshold of 10 while maintaining a consistent relationship with the associated microflora (Table S2). Forward MR analysis was conducted using intestinal flora as the exposure factor and type 2 diabetes as the outcome. We identified 10 types of intestinal flora that have a causal relationship with type 2 diabetes, including the class.Coriobacteriia, order.Coriobacteriales, family.BacteroidalesS24.7group, family.Coriobacteriaceae, genus.Escherichia, genus.Shigella, genus.Ruminococcustorquesgroup, and genus.Victivallis, which are linked to the elevated risk of type 2 diabetes (P < 0.05, odds ratio (OR) > 1). In contrast, family.Veillonellaceae, genus.Howardella, and genus.Romboutsia were linked to the reduced risk of type 2 diabetes (P < 0.05, OR <1) (Table 1, Table S3, Figure 1). The leave‐one sensitivity analysis showed no potential outliers among the IVs for specific bacterial group. No significant heterogeneity was detected in the Cochran Q test of the IVW results (P > 0.05), and no significant horizontal pleiotropy was observed in the MR‐Egger regression analysis (P > 0.05). The sensitivity analysis demonstrated the reliability of the MR findings. Table 1Intestinal flora associated with type 2 diabetesTypeNameMethodnsnp P valueOR (95% CI)Pleiotropy_P valueQ_P valueClassCoriobacteriiaIVW50.021.18 (1.03–1.37)0.160.21OrderCoriobacterialesIVW50.021.18 (1.03–1.37)0.160.21FamilyBacteroidalesS24.7groupIVW501.11 (1.03–1.19)0.730.91FamilyCoriobacteriaceaeIVW50.021.18 (1.03–1.37)0.160.21FamilyVeillonellaceaeIVW90.040.93 (0.86–1)0.90.6GenusEscherichia.ShigellaIVW40.041.13 (1–1.26)0.590.81GenusHowardellaIVW70.020.95 (0.92–0.99)0.960.56GenusRomboutsiaIVW30.010.84 (0.73–0.95)0.910.88GenusRuminococcustorquesgroupIVW50.041.13 (1.01–1.27)0.750.96GenusVictivallisIVW50.021.07 (1.01–1.14)0.680.25CI, confidence interval; IVW, inverse variance weighting; nSNP, the number of single nucleotide polymorphisms; OR, odds ratio. Intestinal flora associated with type 2 diabetes CI, confidence interval; IVW, inverse variance weighting; nSNP, the number of single nucleotide polymorphisms; OR, odds ratio. The causal relationship of intestinal flora and type 2 diabetes. (a–j) Scatter plot of causal relationship between intestinal flora and type 2 diabetes; (k) Forest plot of causal relationship between intestinal flora and type 2 diabetes. CI, confidence interval; OR, odds ratio; nSNP, the number of single nucleotide polymorphisms. To further confirm whether the above 10 kinds of intestinal flora were also affected by type 2 diabetes, the study took type 2 diabetes as the exposure factor and 10 kinds of intestinal flora as the outcome for reverse MR analysis. A total of 8 SNPs were selected as IVs for type 2 diabetes, with all IVs showing F statistics ranging from 10.80 to 31.70, exceeding the threshold of 10 (Table S4). The results indicated that none of the 10 intestinal flora were affected by type 2 diabetes (Table 2, Table S5). In summary, these findings suggest that while these 10 intestinal florae have a causal relationship with type 2 diabetes and conform to three hypotheses, which can be included in further studies. Table 2Reverse mendelian randomization (MR) of intestinal flora and T2DTypeNameMethodnsnp P‐valueOR (95% CI)Pleiotropy_P valQ_P valClassCoriobacteriiaIVW60.130.95 (0.88–1.02)0.290.37OrderCoriobacterialesIVW60.130.95 (0.88–1.02)0.290.37FamilyBacteroidalesS24.7groupIVW60.931 (0.91–1.11)0.670.66FamilyCoriobacteriaceaeIVW60.130.95 (0.88–1.02)0.290.37FamilyVeillonellaceaeIVW60.610.98 (0.91–1.06)0.890.96GenusEscherichia.ShigellaIVW60.870.99 (0.91–1.08)0.420.8GenusHowardellaIVW60.150.9 (0.77–1.04)10.44GenusRomboutsiaIVW60.250.94 (0.85–1.04)0.510.09GenusRuminococcustorquesgroupIVW60.380.97 (0.91–1.04)0.80.87GenusVictivallisIVW50.920.99 (0.84–1.17)0.911CI, confidence interval; IVW, inverse variance weighting; nSNP, the number of single nucleotide polymorphisms; OR, odds ratio. Reverse mendelian randomization (MR) of intestinal flora and T2D CI, confidence interval; IVW, inverse variance weighting; nSNP, the number of single nucleotide polymorphisms; OR, odds ratio. eQTAL MR analysis showed that 228 genes had causal relationship with type 2 diabetes Genes have a complex influence on individual characteristics and traits. In order to lay a foundation for revealing the mechanism of gene influence on type 2 diabetes, FUSION was used to conduct TWAS analysis and identify genes related to type 2 diabetes. The relationship between cis‐SNPs and gene expression was constructed by FUSION, predicting gene expression and correlates with traits, which revealed 1,706 genes associated with type 2 diabetes (Table S6). The TWAS results indicated that the expression levels of CAMK1D, WFS1, SPPL3, CDKAL1, and SLIT1 might be positively correlated with type 2 diabetes (Z score > 1), whereas the expression levels of PABPC4 and BCAR1 may be negatively correlated with type 2 diabetes (Z score < −1) (Figure 2a). Identification of genes causally linked to type 2 diabetes.(a) Transcriptome‐wide association study of type 2 diabetes; (b, c) Screening of intersection genes; (d) The top 20 genes with the lowest P‐value in MR Analysis among intersection genes. CI, confidence interval; MR, mendelian randomization; nSNP, the number of single nucleotide polymorphisms; OR, odds ratio. To identify the genes that had a causal relationship with type 2 diabetes, this study used cis‐eQTL of 1706 type 2 diabetes‐related genes as the exposure factor and type 2 diabetes as the outcome conducted MR analysis. The results indicated that 228 type 2 diabetes‐related genes were causally linked to type 2 diabetes. Among them, 112 were identified (OR < 1) as protective factors and 116 as risk factors (OR > 1) (Table S7). To enhance the accuracy of the findings, we selected genes exhibiting consistent directional effects in both the eQTL MR and TWAS analyses for joint evaluation. The results demonstrated that 110 of the 112 protective factors identified in the eQTL MR analysis were negatively correlated with type 2 diabetes in the TWAS analysis, while 111 of the 116 risk factors showed a positive correlation with type 2 diabetes, indicating a high level of consistency between the MR and TWAS analyses (Figure 2b–d, Table S8). The consistent trend of MR analysis results and TWAS results further verified the TWAS analysis results and strengthened the correlation between type 2 diabetes‐related genes and its underlying mechanisms. To investigate the effects of genes on intestinal flora, this study utilized the 221 genes causally linked to type 2 diabetes as exposure factors and the 10 types of intestinal flora as outcomes to conducting MR analysis. The results showed that 37 causal genes out of 221 genes causally linked to type 2 diabetes were causally linked to at least one of nine intestinal flora, such as NAALADL1 was associated with three types of intestinal flora, and the genus Escherichia.Shigella was associated with 9 causal genes (Figure 3, Table S9). Through the above analysis, genes that have a causal relationship with both intestinal flora and type 2 diabetes were obtained, which proved the mediating role of genes in the intestinal flora affecting the development of type 2 diabetes. Identification of causal genes linking gut microbiota through MR analysis. MR, mendelian randomization; nSNP, the number of single nucleotide polymorphisms; OR, odds ratio (OR >1 indicates that the exposure is a risk factor for the outcome. OR <1 indicates that the exposure is a protective factor for the outcome); CI, confidence interval. To further explore the regulatory relationships among causal genes, intestinal flora, and type 2 diabetes, we utilized Cytoscape software to construct a three‐level regulatory network encompassing genes, intestinal flora, and type 2 diabetes, which contained TMEM206‐genus.Howardella‐type 2 diabetes, TOM1‐genus.Escherichia‐type 2 diabetes and NAALADL1‐order. Coriobacteriales‐type 2 diabetes regulation relationship (Figure S1). Biological function and pathway analysis of 37 causal genes Enrichment analysis was performed to further investigate the biological functions of 37 causal genes. GO annotation revealed that causal genes were mainly linked to endoplasmic reticulum lumen and lysosome, which were implicated in biological processes including tissue homeostasis and extracellular structural organization, primarily through their roles in REDOX enzyme activity and bile acid binding. KEGG enrichment analysis indicated that the causal genes primarily associated with the lysosome, glycosaminoglycan biosynthesis, fatty acid elongation, and other signaling pathways (Figure 4a–d, Table S10). Gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) enrichment analysis of 37 causal genes that exhibit causal relationships with both gut microbiota and type 2 diabetes. (a–c) GO enrichment analysis. (a) Biological process enrichment; (b) Cellular component enrichment; (c) Molecular function enrichment; (d) KEGG enrichment analysis. Identification of 20 differentially expressed causal genes To further analyze the expression of DEGs in the type 2 diabetes transcriptome data, the GSE76894 dataset was differentially analyzed. PCA analysis revealed a significant clustering or separation of type 2 diabetes and control samples in the spatial dimension (Figure 5a). Analysis of gene expression differences demonstrated that there were 7,241 down‐regulated and 6,760 up‐regulated genes in the T2D samples (P < 0.05) (Figure 5b). The expression profiles of causal genes were extracted from the GSE76894 dataset, while the gene expression of LGSN, ZNF613, and FAM153A were not found, so 34 causal genes were selected for further analysis. Immunoinfiltration and drug prediction of differentially expressed causal genes. (a) Principal component analysis (PCA) of control group and type 2 diabetes sample; (b) Identification of differentially expressed genes (DEGs); (c) Expression of 34 causal genes in control and type 2 diabetes samples. (d) Difference of immune infiltration between control and type 2 diabetes samples. (e) Correlation of gene and immune cell in T2D samples. (f) Drug prediction of causal genes. Red circles represent genes and blue arrows represent drugs. ns, P > 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. From the dataset, the expression of 34 causal genes was extracted and analyzed, with comparisons made between type 2 diabetes and control samples using the Wilcoxon test. The results indicated that 20 causal genes were differentially expressed between the two groups, with expression trends aligning with the MR analysis findings—specifically, risk genes were found to be highly expressed in type 2 diabetes samples, while protective genes showed reduced expression in type 2 diabetes samples (Figure 5c). This consistency further supports the reliability of our study results. type 2 diabetes immunoinfiltration analysis and potential drug prediction Studies have shown that the gut microbiota can regulate the immune system and causes at least in part type 2 diabetes 27 . Moreover, chronic inflammation in type 2 diabetes is characterized by the continuous activation of immune cells 28 . Therefore, to investigate the difference in immune infiltration levels of type 2 diabetes and control samples, immune infiltration analysis was conducted using the GSE76894 dataset. The findings showed that immune infiltration levels in samples from type 2 diabetes patients were notably greater than those observed in the control group, such as activated B cell, activated CD4 T cell, CD56bright natural killer cell, and neutrophil cell (Figure 5d). Correlation analysis indicated that BCAT1 exhibited a significantly positive correlation with the infiltration of activated B cells, activated CD4 T cells, activated CD8 T cells, and various other immune cells (cor >0.4, P < 0.05). Conversely, the infiltration of eosinophils, regulatory T cells, and other immune cells showed a negative correlation with BCAT1 (cor < −0.4, P < 0.05) (Figure 5e). This study conducted drug predictions to identify potential drugs for T2D. The results showed that only three genes were predicted to have targeted drugs, with the highest number of drugs targeting BLVRA, including incadronic acid and alendronate sodium (Figure 5f). Identification of type 2 diabetes diagnostic markers using machine learning In order to find the diagnostic marker of type 2 diabetes, the study further screened the differentially expressed causal genes. The machine learning‐LASSO, RF, and SVM‐RFE were employed to identify 20 genes whose expression trends aligned with MR analysis. Specifically, these algorithms selected 12, 5, and 20 potential diagnostic markers, respectively (Figure 6a–g). By overlapping results in a Venn diagram, four potential diagnostic markers were identified: BEND7, BLVRA, C1orf85, and LAMC1 (Figure 6h). Notably, BEND7 and BLVRA were significantly overexpressed in samples with type 2 diabetes, whereas C1orf85 and LAMC1 were significantly underexpressed (Figure 6i). To assess the diagnostic effectiveness of potential diagnostic markers, ROC curves were generated; four genes were found to have high efficiency: 0.822 (BEND7), 0.870 (BLVRA), 0.885 (C1orf85), 0.961 (LAMC1), indicative of high diagnostic potential. A logistic regression model was subsequently constructed, achieving a diagnostic efficiency of 0.998 (Figure 6j). This indicated that the diagnostic model based on four diagnostic markers had high diagnostic value for type 2 diabetes. Screening and validation of diagnostic markers. (a, b) Regression coefficient path graph and cross verification curve in least absolute shrinkage and selection operator (LASSO); (c) curves of the predicted truth value and error value of gene in support vector machine recursive feature elimination (SVM‐RFE); (d–g) feature importance recognition based on random forest. OOB, out of bag; (h) the intersection of diagnostic markers obtained from the three algorithms; (i) expression of four diagnostic markers in the training set. (j) ROC curve of the diagnostic model in the training set. AUC, area under the curve; (k) expression of the diagnostic markers in the validation set. (l) receiver operating characteristic (ROC) curve of the diagnostic model in the validation set. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. To verify the universality of the diagnostic model constructed, we further validated the diagnostic efficacy of the four diagnostic markers using the GSE164416 validation set, which demonstrated that the expression levels of these genes were consistent with those observed in the training set (Figure 6k). This alignment reinforces the robustness and reliability of our diagnostic model across different datasets, suggesting that these markers could be universally applicable for diagnosis in diverse populations. Further analysis may include exploring their predictive power and specificity in various clinical scenarios. Additionally, the AUC values of the ROC curves of the four genes were 0.848 (BEND7), 0.813 (BLVRA), 0.751 (C1orf85), 0.701 (LAMC1), ROC curves confirmed that the diagnostic model exhibited high accuracy (Figure 6l). Exploration of the expression of diagnostic markers in different cells The above machine learning analysis identified four diagnostic markers: BEND7, BLVRA, C1orf85, and LAMC1, and their diagnostic efficacy was verified in an external dataset. To further explore the cell type‐specific roles of these four diagnostic markers in the pathogenesis of type 2 diabetes and to verify whether the expression differences are due to abnormalities in specific pancreatic cells, we analyzed the expression patterns of the markers in different pancreatic cell types using the GSE221156 single‐cell dataset. The single‐cell dataset underwent quality control, resulting in the identification of 27,937 genes from 167,585 cells. Violin plots illustrated the number of genes detected in each cell, along with the total count and percentage of mitochondrial genes (Figure S2A). These visualizations provided insights into the overall quality and diversity of the cellular transcriptomes. The distribution patterns indicated variations in gene expression levels across cells, highlighting potential differences in cellular states or conditions. Correlation analysis showed that the correlation coefficient between nCount and nFeature was 0.91, indicating a significant positive correlation (Figure S2B). This strong relationship suggests that as the total number of unique genes detected nFeature increases, the overall count of transcripts nCount also tends to rise. Such a finding may imply that cells with higher transcriptional activity are more likely to express a greater variety of genes, reflecting enhanced cellular complexity or metabolic activity. Figure S2C highlighted 2,000 highly variable genes, while PCA analysis identified a total of 20 principal components (Figure S2D). The first 20 principal components were selected for subsequent analysis. According to t‐Distributed Stochastic Neighbor Embedding (tSNE) and cell type notes, all cells were divided into 9 categories: 56,630 alpha cells, 54,435 beta cells, 3,166 gamma cells, 7,816 delta cells, 11,497 acinar cells, 11,323 stellate cells, 16,383 ductal cells, 3,948 endothelial cells, and 2,387 immune cells (Figure 7a–c). We then analyzed DEGs within each cell type as well as the number and proportion of each cell type across patients (Figure 7d,e). Comparative analysis of the proportions of different cell types between samples with type 2 diabetes and control samples revealed a significantly higher content of alpha, ductal, and stellate cells in the samples with type 2 diabetes, while beta cells were significantly reduced (Figure 7f). Additionally, we compared the expression of four selected diagnostic markers across different cell types. The results indicated that the expression of BEND7 was inhibited in gamma cells of samples with type 2 diabetes, while C1orf85 was specifically highly expressed in ductal cells of samples with type 2 diabetes(Figure 7e). Expression of cell types and diagnostic markers in type 2 diabetes. (a) Clustering of cells. tSNE: t‐distributed stochastic neighbor embedding; (b) results of cell annotation; (c) markers for the annotated cell types; (d) volcano map of DEGs in different cells; (e) the number and proportion of each cell type in each sample; (f) difference in cell proportion between type 2 diabetes and control samples. (g) Diagnostic gene expression in different cell types. ns, P > 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. DISCUSSION Multiple studies have demonstrated that gut microbiota imbalance is a significant risk factor for type 2 diabetes. For instance, reduced levels of beneficial bacteria such as Lactobacillus and Bifidobacterium, along with increased levels of harmful bacteria like Bacteroidota and Clostridium, are associated with higher type 2 diabetes risk 29 . To further investigate the potential causal relationship between gut microbiota and type 2 diabetes, this study employed MR analysis to identify 10 gut microbiota species significantly linked to type 2 diabetes, and 37 genes associated with both nine gut microbiota types and type 2 diabetes. Furthermore, 4 biomarkers—BEND7, BLVRA, C1orf85, and LAMC1—potentially involved in T2D progression were screened out through machine learning algorithms. The results revealed that class. Coriobacteriia, order. Coriobacteriales, family. Bacteroidales S24.7 group, family. Coriobacteriaceae, genus. Escherichia.Shigella, and genus. Ruminococcus torques group were associated with increased risk of type 2 diabetes. These microbial communities span multiple taxonomic levels, from phyla to genera, and may possess unique biological characteristics and functions within the gut microbiota. These distinctive features could potentially contribute to the elevated risk of developing type 2 diabetes. Coriobacteriaceae, as a significant component of the gut microbiota, may significantly influence host metabolic status through changes in abundance. Studies have shown that gut microbiota dysbiosis can lead to insulin resistance and exacerbated inflammatory responses, which may be closely associated with the pathogenesis of T2D 30 , 31 . During chronic inflammatory diseases, the population of Coriobacteriaceae typically decreases, and its potential probiotic effects make it a promising therapeutic target 32 . Moreover, alterations in Coriobacteriaceae may be closely linked to the development of T2D complications. Research reveals that compared to healthy individuals, diabetic patients exhibit significant differences in both diversity and composition of gut microbiota, which may affect disease progression and complication occurrence 33 . The reduction of Coriobacteriaceae may play a crucial role in this process. Additionally, genus. Escherichia.Shigellas a common intestinal bacterial genus, some strains possess pathogenic properties that can trigger inflammatory responses in the gut. Escherichia.Shigella has been confirmed to be closely associated with the development of T2D 34 . It produces inflammatory components such as lipopolysaccharides and peptidoglycan, causing intestinal and systemic inflammation, along with subsequent insulin resistance‐key drivers of T2D 35 . In T2D models, elevated levels of this bacterium and reduced abundance of its beneficial counterparts have been shown to alleviate symptoms in mice with T2D 36 . Our study further identified Escherichia.Shigella as a risk factor for T2D, highlighting its significance in disease management. Additionally, Escherichia.Shigella has been strongly associated with T2D‐related complications including diabetic nephropathy 37 and chronic diabetic foot infections 38 . These findings collectively demonstrated Escherichia.Shigella's critical role in diabetes and its associated disorders. While the exact role of genus.Ruminococcustorquesgroup and genus.Victivallis in T2D pathogenesis remains poorly understood, their presence and relative abundance may disrupt the intestinal microecological balance when viewed holistically. This imbalance could indirectly impair host metabolic health by affecting the growth and metabolism of beneficial microbes, or through interactions with the immune system. Ultimately, these changes may heighten the risk of developing T2D. In contrast, family.Veillonellaceae, genus.Howardella and genus.Romboutsia have been found to be associated with a reduced risk of developing T2D. While the exact mechanisms of family.Veillonellaceae in intestinal insulin function remain unclear, studies have demonstrated that its bacterial analogs effectively reduce toxic bile acids 39 . As beneficial gut microbiota, genus.Romboutsia are more abundant in normal mice than in diabetic counterparts 40 . Research indicates that Romboutsia‐driven microbiomes, characterized by low bacterial diversity and high primary bile acid levels, are linked to adipogenesis‐related obesity 41 . However, current research on these bacteria in T2D remains limited, necessitating further investigation to determine their critical roles in the disease. In recent years, with the development of metagenomic sequencing technology, researchers have been able to analyze the relationship between the gut microbiota and T2D at the strain level 42 . The human gut microbiome contains 100 trillion bacteria 43 . Large‐scale cross‐cohort metagenomic analyses involving more than 8,000 individuals have demonstrated that strain‐level phylogenetic diversity can explain individual variations in T2D susceptibility; typical examples include Eubacterium rectale and Prevotella copri, whose specific strain gene repertoires are closely involved in branched‐chain amino acid biosynthesis, quorum sensing, and horizontal gene transfer, thereby mediating metabolic dysregulation and chronic low‐grade inflammation in patients with type of patients 42 . Accumulating evidence indicates that even within the same bacterial species, distinct strains exhibit substantial genomic and functional heterogeneity, leading to divergent effects on host glucose homeostasis, inflammatory response, and insulin sensitivity 43 . Enrichment analysis revealed that 37 T2D‐related genes were primarily enriched in signaling pathways such as lysosomes, glycosaminoglycan biosynthesis, and fatty acid elongation. These genes were involved in redox enzyme activity, bile acid conjugation, and other biological processes related to tissue homeostasis and extracellular structural organization. Abnormal expression of endoplasmic reticulum (ER) cavity proteins associated with protein folding may lead to increased misfolded proteins, triggering ER stress. This stress can further disrupt normal insulin signaling pathways, reducing cellular insulin sensitivity and ultimately contributing to the development of type 2 diabetes 44 . Research by Cheng demonstrated that lipoprotein lipase (LPL) degradation is regulated by multiple related proteins and occurs both intracellularly and extracellularly, both involving lysosomes. LPL degradation is as crucial as biosynthesis and closely associated with T2D. Promoting macrophage‐derived LPL degradation can prevent foam cell formation and reduce inflammatory factor expression, while inhibiting LPL degradation enhances its stability and enzymatic activity to lower triglyceride levels and inhibit atherosclerosis (a common complication of T2D) 45 . Previous studies also showed reduced levels of lysosomal function‐related markers Lamp1 and Lamp2 in the hippocampus of diabetic rats, suggesting that diabetes may impair lysosomal function 46 . In summary, lysosomes participate in influencing the onset, progression, and complications of type 2 diabetes through multiple pathways. Additionally, Yoshida et al. identified ELOVL6 as the rate‐limiting enzyme in long‐chain fatty acid synthesis and elongation 47 . Epidemiological studies demonstrate that a high‐fat, high‐carbohydrate diet induces excessive fatty acid accumulation in the body, triggering tissue cell lipid toxicity (characterized by chronic low‐grade inflammation, oxidative stress, and functional impairment)‐conditions strongly associated with obesity and insulin resistance 48 . This suggests that ELOVL6 may disrupt lipid metabolism through oxidative stress, thereby influencing type 2 diabetes. These findings indicate that the identified gene exerts multifaceted effects in T2D through multiple signaling pathways, collectively orchestrating the development and progression of this metabolic disorder. Through machine learning, we identified 4 biomarkers for diagnosis: BEND7, BLVRA, C1orf85, and LAMC1. Among these, BEND7 (BEN domain‐containing protein 7) belongs to the BEN domain transcription factor family. Its α‐helix BE module mediates protein‐DNA and protein–protein interactions, thereby regulating chromatin architecture and gene transcription 49 . Although this gene has been found to show significant correlation with carbohydrate intake or blood glucose levels in Indian populations, no such association has been observed in other groups 50 . Our study, for the first time, demonstrated a significant causal relationship between BEND7 and T2D, laying the groundwork for future research and providing insights into disease risk reduction. BLVRA (biliverdin reductase A), a major subtype of biliverdin reductase (BVR), possesses both antioxidant and cellular transport functions 51 . It facilitates the conversion of biliverdin to bilirubin and acts as an intracellular transporter to regulate tumor cell growth and proliferation. In various cancers, including liver, lung, breast, and esophageal cancers, BLVRA exhibits overexpression that promotes tumor development 51 . While its overexpression affects thousands of genes, signaling pathways, apoptosis, and immune molecules, its role in T2D remains unexplored. Our study revealed for the first time that BLVRA was associated with T2D. LAMC1 (gamma‐1 laminin subunit), a glomerular filtration barrier‐related factor, plays a crucial role in basement membrane formation and function. Previous studies show elevated LAMC1 expression in diabetic mice, positively correlated with urinary protein excretion rates 52 . Our combined MR analysis and expression results confirmed: BEND7 and BLVRA are risk factors, while C1orf85 and LAMC1 are protective factors. Significantly, BEND7/BLVRA showed high expression in disease samples, whereas C1orf85/LAMC1 demonstrated low expression. The consistent trend between MR analysis and expression enhanced the reliability of results. However, systematic mechanistic studies of these four genes in T2D remain lacking, necessitating further research to identify novel therapeutic targets for precision medicine. Our drug prediction results suggested that BLVRA may be a potential therapeutic target for T2D, as it is predicted to interact with drugs such as incadronic acid and alendronate sodium. However, these findings are based on computational predictions from the DGIdb database and require further validation in experimental models and clinical settings. Nevertheless, they provide testable hypotheses for future drug repurposing or development efforts in T2D management. The latest research indicates that there is a communication axis between the gut and the pancreas, enabling the gut to strongly influence an animal's glucose metabolism and energy homeostasis 53 . The ‘gut‐pancreas axis’ is an important endocrine signaling axis that regulates pancreatic function through the dialog between the gut microbiome ecology and endocrine metabolism 53 . The composition of the gut microbiota can affect pancreatic function. When the gut microbiota is imbalanced, pro‐inflammatory factors such as lipopolysaccharide (LPS) enter the circulation system, triggering systemic low‐grade inflammation, which in turn promotes the activation of stellate cells and the proliferation of ductal cells 54 . At the same time, microbial metabolites can affect the function and survival of α cells and β cells by regulating glucagon‐like peptide‐1 (GLP‐1) 53 . Through scRNA‐seq analysis, it was found that the content of alpha cells, ductal cells and stellate cells in disease samples was significantly higher than that of the control group, while the content of beta cells was significantly lower than that of the control group. Increased alpha cells may disrupt glucagon secretion, leading to elevated blood glucose levels. This phenomenon is likely associated with pancreatic islet dysfunction in patients with type 2 diabetes. As previous studies have shown, islet dysfunction can interfere with cellular secretion regulation mechanisms and affect glucose metabolism balance 55 . The elevated ductal cell content in samples with type 2 diabetes, given their importance in pancreatic exocrine function and islet microenvironment maintenance, may alter intercellular interaction patterns within the islet, indirectly affecting insulin secretion and action. Research has confirmed that altered islet microenvironment significantly impacts insulin function 56 . Stellate cells play a critical role in pancreatic fibrosis and inflammatory responses. In the context of type 2 diabetes, activation of stellate cells may promote pancreatic fibrosis and further impair islet function 57 , 58 . Beta cells, as the primary insulin‐secreting cells, show a significant reduction in number among patients with type 2 diabetes, directly causing insufficient insulin secretion. This constitutes one of the core mechanisms of type to diabetes development. Various factors such as genetics, autoimmune responses, and metabolic stress can lead to impaired beta cell function or even apoptosis 59 . In conclusion, these single‐cell analysis results provide important clues for understanding the pathogenesis of type to diabetes, exploring new therapeutic targets and developing effective therapeutic strategies. Although recent studies support the existence of a gut‐islet or gut‐pancreas axis, the currently cited articles mainly provide evidence that gut microbiota and gut‐derived metabolites may influence pancreatic islet function, glucose metabolism, inflammatory responses, and pancreatic microenvironment remodeling. They do not directly demonstrate an association between gut microbiota and pancreatic gamma cells, nor do they specifically describe BEND7‐expressing gamma cells. Therefore, the reduced expression of BEND7 in gamma cells observed in our scRNA‐seq analysis should be interpreted as a cell type‐specific finding rather than direct evidence that gut microbiota regulates BEND7 expression in pancreatic gamma cells. In this study, the potential relationship between gut microbiota and BEND7 was inferred from the MR‐based microbiota‐gene‐type 2 diabetes analytical framework, whereas the gamma cell‐specific expression pattern of BEND7 was identified from the independent single‐cell dataset. Further experimental studies are required to determine whether gut microbiota or microbiota‐derived metabolites can directly affect BEND7 expression in pancreatic gamma cells. This study using two‐sample MR analysis to detect causal associations between gut microbiota and type 2 diabetes. Utilizing these microbiota, it systematically identified biomarkers linked to type 2 diabetes progression, offering new potential targets for type 2 diabetes diagnosis. However, this study also has the following limitations: First, the study primarily focuses on European populations, and its conclusions may not fully represent type 2 diabetes patient cohorts with diverse geographic, ethnic, and genetic backgrounds, limiting generalizability. Future research should expand sample coverage and evaluate gut microbiota‐type 2 diabetes associations across different ethnicities and geographic regions. Second, since we only used publicly available aggregated‐level genome‐wide association analysis data, individual dietary information at the participant level could not be obtained. Therefore, this study did not control for confounding dietary variables. In the future, large‐scale prospective cohort studies need to be conducted and detailed personal dietary records need to be collected to further validate our research results and eliminate potential environmental interference factors. Third, using unadjusted P‐values as screening criteria helps identify biological signals but may elevate false‐positive risk. These results warrant further experimental verification. Fourth, the identified potential diagnostic biomarkers require further validation in larger‐scale, multi‐center independent studies to ensure accuracy and clinical applicability. Fifth, although our integrated analysis suggested a potential link among gut microbiota, type 2 diabetes‐related genes, and pancreatic cell type‐specific expression patterns, direct experimental evidence connecting gut microbiota with BEND7‐expressing pancreatic gamma cells remains unavailable. Therefore, this finding should be considered hypothesis‐generating and requires further validation in mechanistic studies. CONCLUSION This study employed bidirectional two‐sample MR analysis to investigate the causal relationship between gut microbiota and type 2 diabetes. Key genes demonstrating causal associations with both gut microbiota and type 2 diabetes were also identified. Machine learning methods including LASSO, SVM‐RFE, and RF were applied, combined with gene expression analysis and ROC curve evaluation, to identify diagnostic biomarkers. Single‐cell data analysis revealed distinct expression patterns of these biomarkers across different cell types, providing insights into their pathogenic mechanisms. The research not only establishes a theoretical foundation for understanding type 2 diabetes pathogenesis and disease progression, but also provides reliable molecular markers for pre‐symptomatic diagnosis. Furthermore, it injects new momentum into precision medicine development. FUNDING This research was funded by the Sanming Project of Medicine in Shenzhen (grant number: SZZYSM202411016). DISCLOSURE The authors declare no conflict of interest. Approval of the research protocol: N/A. Informed consent: N/A. Approval date of Registry and the Registration No. of the study/trial: N/A. Animal studies: N/A. ETHICS STATEMENT This study was based entirely on publicly available summary‐level GWAS, eQTL, transcriptomic, and single‐cell RNA sequencing datasets. No new human participants, human biological specimens, or animal experiments were involved in this study; therefore, additional institutional ethics approval and informed consent were not required. Ethical approval and informed consent for the original datasets were addressed in the original studies. AUTHOR CONTRIBUTIONS Yu‐yang Chen: Writing—original draft, software, project administration, methodology, investigation, data curation, and conceptualization. Yu‐yang Chen: Writing—original draft, methodology, and formal analysis. Yi‐fan Shi: Validation, Software, and formal analysis. Xin‐yue Zhang: Validation, software, and formal analysis. Shi‐mao Zhang: Validation and investigation. Jing‐yue Zhang: Writing—review and editing, supervision, project administration, and data curation. Heng‐xia Zhao, De‐liang Liu: Writing—review and editing, supervision, resources, project administration, funding acquisition, data curation, conceptualization. Supporting information Figure S1. Construction of gene‐intestinal flora‐type 2 diabetes regulatory network. Purple indicates genes, orange indicates intestinal flora, and green indicates type 2 diabetes. Figure S2. Single cell data quality control. (A) The number of genes detected, total count and percentage of mitochondrial genes in each cell in each sample after quality control; (B) Correlation of gene number, number of cells, and percentage of mitochondrial genes; (C) Scatter plot of the top 2,000 highly variable genes; (D) Principal component analysis of cells and show the first 20 PCs. Table S1. The details of data used in the study. Table S2. The instrumental variables (IVs) of the intestinal flora. Table S3. The Mendelian randomization (MR) results of the intestinal flora and type 2 diabetes. Table S4. The IVs of the type 2 diabetes. Table S5. The MR results of the type 2 diabetes and intestinal flora. Table S6. The TWAS results of the type 2 diabetes (TWAS, P < 0.05). Table S7. The cis‐eQTL MR results of the type 2 diabetes. Table S8. The overlap result of the TWAS analysis and MR analysis in the type 2 diabetes. Table S9. The cis‐eQTL MR results of the intestinal flora. Table S10. The results of the enrichment analysis (P < 0.05). ACKNOWLEDGMENTS We would like to sincerely thank the authors for their scientific contribution. Contributor Information Heng‐xia Zhao, Email: 2073558282@qq.com. De‐liang Liu, Email: ldl2580@gzucm.edu.cn. DATA AVAILABILITY STATEMENT All data used in this study are publicly available. The GWAS summary statistics for type 2 diabetes were obtained from the FinnGen database (https://r10.finngen.fi/pheno/T2D). Genetic associations for gut microbiota were obtained from the MiBioGen consortium (https://mibiogen.gcc.rug.nl/menu/main/home/). The blood cis‐eQTL data were obtained from the eQTLGen database (https://www.eqtlgen.org/phase1.html). The scRNA‐seq dataset (GSE221156), the training dataset (GSE76894), and the validation dataset (GSE164416) were retrieved from the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/). No new data were generated in this study. REFERENCES Associated Data Supplementary Materials Figure S1. Construction of gene‐intestinal flora‐type 2 diabetes regulatory network. Purple indicates genes, orange indicates intestinal flora, and green indicates type 2 diabetes. Figure S2. Single cell data quality control. (A) The number of genes detected, total count and percentage of mitochondrial genes in each cell in each sample after quality control; (B) Correlation of gene number, number of cells, and percentage of mitochondrial genes; (C) Scatter plot of the top 2,000 highly variable genes; (D) Principal component analysis of cells and show the first 20 PCs. Table S1. The details of data used in the study. Table S2. The instrumental variables (IVs) of the intestinal flora. Table S3. The Mendelian randomization (MR) results of the intestinal flora and type 2 diabetes. Table S4. The IVs of the type 2 diabetes. Table S5. The MR results of the type 2 diabetes and intestinal flora. Table S6. The TWAS results of the type 2 diabetes (TWAS, P < 0.05). Table S7. The cis‐eQTL MR results of the type 2 diabetes. Table S8. The overlap result of the TWAS analysis and MR analysis in the type 2 diabetes. Table S9. The cis‐eQTL MR results of the intestinal flora. Table S10. The results of the enrichment analysis (P < 0.05). Data Availability Statement All data used in this study are publicly available. The GWAS summary statistics for type 2 diabetes were obtained from the FinnGen database (https://r10.finngen.fi/pheno/T2D). Genetic associations for gut microbiota were obtained from the MiBioGen consortium (https://mibiogen.gcc.rug.nl/menu/main/home/). The blood cis‐eQTL data were obtained from the eQTLGen database (https://www.eqtlgen.org/phase1.html). The scRNA‐seq dataset (GSE221156), the training dataset (GSE76894), and the validation dataset (GSE164416) were retrieved from the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/). No new data were generated in this study.