Identifying Common Diagnostic Biomarkers and Therapeutic Targets between COPD and Sepsis: A Bioinformatics and Machine Learning Approach

Abstract

Background: Evidence suggests a bidirectional association between chronic obstructive pulmonary disease (COPD) and sepsis, but the underlying mechanisms remain unclear. This study aimed to explore shared diagnostic genes, potential mechanisms, and the role of immune cells in the COPD-sepsis relationship using Mendelian randomization (MR) and bioinformatics approaches, while also identifying potential therapeutic drugs.

Methods: Two-sample MR analysis was performed using genome-wide association data to assess genetically predicted COPD and sepsis. Immune cell-mediated effects were quantified using a two-way two-sample MR analysis. Differential expression gene (DEG) analysis and weighted gene co-expression network analysis (WGCNA) were used to identify common genes. Functional enrichment analyses were conducted to explore the biological roles of these genes. LASSO and SVM-RFE algorithms identified shared diagnostic genes, which were evaluated using receiver operating characteristic (ROC) curves. Immune cell infiltration was analyzed with CIBERSORT, while transcription factor (TF) and miRNA networks were constructed using NetworkAnalyst. Drug predictions were made using DSigDB, and molecular docking validated potential drugs.

Results: Three immune cell types were identified as mediators between COPD and sepsis, with genetically predicted effects mediated by these cells at rates of 6.5%, 12.8%, and 3.9%. A total of 33 overlapping genes were identified, and AIM2 and RNF125 were highlighted as key diagnostic genes. Immune infiltration analysis revealed dysregulated monocyte, macrophage, plasma, and dendritic cells. Regulatory network analysis identified nine key co-regulators. Ten potential drug targets were identified, with seven validated via molecular docking.

Conclusion: AIM2 and RNF125 may serve as diagnostic biomarkers, and identified immune cell subsets could mediate the COPD-sepsis connection, offering insights into potential therapeutic targets.

Keywords: Mendelian randomization; chronic obstructive pulmonary disease; co-diagnostic genes; comprehensive bioinformatics analysis; immune cells; machine learning; molecular docking; predictive drugs; sepsis.

Figure 1

The work flow chart of this study.

Figure 2

The associated charts studied in this study. (A) Bidirectional MR analysis of chronic obstructive pulmonary disease (COPD) and sepsis. c is the total effect of exposure and sepsis as a result of gene-predicted COPD. d is the total effect of genetic prediction of sepsis as a result of exposure and COPD as a result. (B) The total effect was decomposed into: (i) indirect effect using two-step method (a is the effect of COPD on immune cells, b is the effect of immune cells on sepsis) and product method (a×b); (ii) Direct effect (c′ = c−a×b). The proportion mediated by immune cells is the indirect effect divided by the total effect.

Figure 3

Forest plot illustrating the causal relationship of COPD with sepsis. (A) With COPD as the exposure factor and sepsis as the outcome factor, using the IVW, MR Egger, and weighted median methods to study the causal relationship between the two. (B) With sepsis as the exposure factor and COPD as the outcome factor, using the IVW, MR Egger, and weighted median methods to study the causal relationship between the two.

Figure 4

Schematic illustration of the mediating role of immune cell characteristics. (A) The immune cell CD33bright HLA DR+ CD14- %CD33bright HLA DR+ play a mediating role between COPD and sepsis and its indirect effect values. (B) The immune cell BAFF-R on IgD- CD38bright play a mediating role between COPD and sepsis and its indirect effect values. (C) The immune cell CD45 on CD33- HLA DR+ play a mediating role between COPD and sepsis and its indirect effect values.

Figure 5

Identification of differentially expressed genes (DEGs). (A and B) Volcano plots of all DEGs in COPD (GSE5058) and sepsis (GSE95233), respectively. (C and D) Heatmaps of all DEGs in COPD (GSE5058) and sepsis (GSE95233), respectively. (E) Venn diagram of overlapping DEGs between COPD and sepsis.

Figure 6

Functional enrichment analysis of common DEGs between COPD and sepsis. GO analysis results of the common DEGs: displaying the top 5 biological process (BP) terms, top 5 cellular component (CC) terms, and top 5 molecular function (MF) terms. Two KEGG pathways associated with the common DEGs.

Figure 7

Construction of WGCNA networks and identification of key modules. (A and B) Selection of soft thresholds for COPD (GSE5058) and sepsis (GSE95233), respectively. (C and D) Hierarchical clustering trees of co-expression gene clusters for COPD (GSE5058) and sepsis (GSE95233), respectively. (E and F) Heatmaps showing the correlation of each module with clinical characteristics in COPD (GSE5058) and sepsis (GSE95233), respectively. (G) Scatter plot of module membership versus gene significance for the Turquoise module in COPD (GSE5058). (H) Scatter plot of module membership versus gene significance for the Blue module in sepsis (GSE95233). (I) Venn diagram of overlapping genes between COPD and sepsis identified through WGCNA analysis.

Figure 8

Functional enrichment analysis based on disease-related common genes. GO classifications obtained from WGCNA analysis between COPD and sepsis, displaying the top 5 terms for biological process (BP), molecular function (MF), and cellular component (CC). KEGG enrichment analysis results obtained from WGCNA analysis between COPD and sepsis, showing the top 5 pathways.

Figure 9

Machine learning for screening candidate biomarkers. (A) Venn diagram showing the intersection of common genes obtained from WGCNA and DEGs. (B and C) Five genes identified as the most suitable for COPD diagnosis based on the Lasso regression algorithm at the minimum binomial deviance. (D) The top 26 genes with the minimum error and highest accuracy in COPD selected based on SVM-RFE. (E) Venn diagram of intersecting genes obtained from two machine learning algorithms in COPD. (F and G) Eleven genes identified as the most suitable for diagnosing sepsis based on the Lasso regression algorithm at the minimum binomial deviance. (H) The top 21 genes with the minimum error and highest accuracy in sepsis selected based on SVM-RFE. (I) Venn diagram of intersecting genes obtained from two machine learning algorithms in sepsis. (J) Venn diagram showing the overlap of final candidate biomarkers between COPD and sepsis.

Figure 10

Assessment of the diagnostic value of candidate biomarkers in the discovery dataset. (A and B) Expression differences of two shared genes in the COPD (GSE5058) discovery dataset. (C and D) Expression differences of two shared genes in the sepsis (GSE95233) discovery dataset. (E and F) ROC curves for the two shared genes in the COPD (GSE5058) discovery dataset. (G and H) ROC curves for the two shared genes in the sepsis (GSE95233) discovery dataset.

Figure 11

Assessment of the diagnostic value of candidate biomarkers in the validation dataset. (A and B) Expression differences of two shared genes in the COPD (GSE8545) validation dataset. (C and D) Expression differences of two shared genes in the sepsis (GSE57065) validation dataset. (E and F) ROC curves for the two shared genes in the COPD (GSE8545) validation dataset. (G and H) ROC curves for the two shared genes in the sepsis (GSE57065) validation dataset.

Figure 12

Analysis of immune cell infiltration. (A–D) Percentages of 22 immune cells identified by the CIBERSORT algorithm in the COPD dataset (GSE5058) and sepsis dataset (GSE95233). (E and F) Box plots showing the proportions of immune cells in COPD and sepsis patients versus controls. (G and H) Correlation of AIM2 (G) and RNF125 (H) with infiltrating immune cells in COPD and normal samples. (I and J) Correlation of AIM2 (I) and RNF125 (J) with infiltrating immune cells in sepsis and normal samples.

Figure 13

Gene-miRNA network of Tfs genes and two shared genes. (A) Regulatory network of Tfs genes for the two shared genes. (B) Gene-miRNA regulatory network for the two shared genes.

Figure 14

Molecular docking diagrams of seven drugs. (A) The molecular docking results of drug Aizen uranine and gene AIM2. (B) The molecular docking results of drug Isotretinoin and gene AlM2. (C) The molecular docking results of drug Benfotiamine and gene RNF125. (D) The molecular docking results of drug Carbamazepine and gene RNF125. (E) The molecular docking results of drug lohexol and gene RNF125. (F) The molecular docking results of drug Metoclopramide and gene RNF125. (G) The molecular docking results of drug MelOx and gene RNF125.