Identification of overlapping genes, enrichment analysis, and PPI network construction
From the GSE54837 dataset, 3,371 DEGs were identified, including 1,675 upregulated and 1,696 downregulated DEGs. The top 10 genes with the most significant up- and downregulation are shown in Figure 1A. Hierarchical clustering of the GSE54837 data was performed (Supplementary Figure 1A), and a soft-thresholding power of 10 was applied to ensure scale-free network topology (Figure 1B). Gene co-expression modules were constructed using the dynamic tree cut method with a minimum module size of 50 genes, and each module was assigned a distinct color (Supplementary Figure 1B). Modules with eigengene correlations > 0.75 were subsequently merged (Supplementary Figure 1C, Figure 1C), resulting in 14 distinct modules. Based on Pearson correlation analysis between module eigengenes and clinical characteristics, the MEsalmon module (comprising 5,226 genes) exhibited the most significant positive correlation with COPD (r = 0.35, p = 7 x 10⁻8, Figure 1D). Venn analysis identified 160 overlapping genes among the 3,371 DEGs, 5,226 MEsalmon module genes, and 2,118 ac4C-RGs (Figure 1E). Functional enrichment analysis of these 160 genes showed that significant GO terms included single-stranded RNA binding, regulation of mRNA metabolic process, and RIG-I signaling pathway (Figure 1F). Additionally, KEGG analysis demonstrated that these genes were primarily enriched in Fc gamma R-mediated phagocytosis, the mRNA surveillance pathway, and focal adhesion (Figure 1G). The PPI network of overlapping genes contained 118 nodes and 196 edges (Figure 1H).
Identification of six key genes in COPD
To further identify potential key genes among the 160 overlapping candidates, three machine learning algorithms were applied. LASSO regression was first applied, with cross-validation used to determine the optimal penalty parameter (λ) ≈ 0.091 (Figure 2A). The coefficient profile plot indicated that 17 genes were retained at the optimal λ value (Figure 2B). XGBoost analysis identified the top 30 genes with the highest gain, among which PTRF, WBP11, and LDOC1L exhibited high predictive value (Figure 2C). The RF algorithm similarly ranked the top 30 genes according to their Gini importance scores, with PTRF, RFX5, and PRKCDBP among the most predictive (Figure 2D). Intersection analysis of genes selected by the three methods identified six key overlapping genes: PTRF, PRKCDBP, UPP1, TOR3A, FAM168B, and B4GALT2 (Figure 2E).
Diagnostic model construction and expression analysis of key genes
Using 70% of the GSE54837 dataset as the training set, a logistic regression model was established incorporating the six key genes. ROC curve analysis indicated moderate diagnostic performance, with AUCs of 0.766 (95% CI: 0.691–0.8417), 0.759 (95% CI: 0.6368–0.8817), and 0.723 (95% CI: 0.6085–0.8596) for the training, internal test, and external validation sets, respectively (Figure 3A–C). Calibration analysis confirmed high reliability, and DCA indicated clear net clinical benefit across a wide range of threshold probabilities in both training (Figure 3D–E) and validation sets (Figure 3F–G). A nomogram was constructed to visualize the contribution of each gene and facilitate individualized risk estimation (Figure 3H). Expression analysis revealed that B4GALT2, PRKCDBP, and UPP1 were significantly upregulated in COPD samples, whereas FAM168B, PTRF, and TOR3A were downregulated (Figure 3I).
Regulatory network and functional analysis of key genes in COPD
A functional interaction network comprising the top 20 genes most related to the identified hub genes was constructed using GeneMANIA analysis (Figure 4A). Physical interactions accounted for the majority of connections, followed by co-expression correlations and shared protein domains. Functional annotation indicated significant enrichment in processes such as nucleobase-containing small molecule catabolic process, nucleoside catabolic process, and plasma membrane raft. Post-transcriptional regulation was investigated by intersecting miRNA predictions from the DIANA-microT and miRanda databases, identifying eight overlapping miRNAs (Figure 4B). A lncRNA-miRNA-mRNA regulatory axis was subsequently constructed. According to the Sankey diagram, two of the identified miRNAs, both associated with the regulation of FAM168B, were predicted to be targeted by seven lncRNAs; no such regulatory interactions were identified for the remaining five hub genes (Figure 4C). Transcriptional regulation was further explored using the ChEA3 platform, which predicted upstream transcription factors (TFs) for B4GALT2, UPP1, FAM168B, and TOR3A. The top ten TFs for each gene were selected to construct a TF-target regulatory network (Figure 4D). GSEA was performed to investigate the biological functions of the six key genes. UPP1 was significantly enriched in biological processes such as diacylglycerol metabolic process and purine nucleoside triphosphate biosynthetic process, as well as pathways including proteasome and metabolism of xenobiotics by cytochrome P450 (Figure 4E–F). Enrichment results for the remaining five key genes are provided in Supplementary Figure 2A–J.
Immune infiltration of UPP1 and prediction of drug targets in COPD
Immune infiltration levels of 28 immune cell types were evaluated in control and COPD groups using the ssGSEA algorithm. In COPD patients, memory B cells, myeloid-derived suppressor cells, and activated dendritic cells showed significantly higher enrichment scores. In contrast, type 1 T helper cells, activated B cells, and immature B cells showed significantly lower enrichment scores (Figure 5A). It should be noted that ssGSEA provides relative immune cell enrichment estimates based on transcriptomic data rather than direct measurements of immune cell proportions. Correlation analysis revealed that the six key genes exhibited differential association patterns with immune cell subsets. Specifically, UPP1, PRKCDBP, and B4GALT2 were positively correlated with the infiltration levels of memory B cells, activated dendritic cells, and myeloid-derived suppressor cells (Spearman ρ > 0.4, p < 0.05), while PTRF, TOR3A, and FAM168B showed negative correlations with type 1 T helper cells and activated B cells (Spearman ρ < −0.3, p < 0.05). The complete correlation matrix is presented in the heatmap (Figure 5B). Drug prediction analysis identified UPP1 as the only gene among the six candidates with predicted interactions with small molecules. Three compounds, including fluorouracil, capecitabine, and 5-benzylacyclouridine, were identified from the database as potential UPP1-interacting compounds (Figure 5C). These compounds are primarily used in oncology or experimental settings, and their relevance to COPD requires further investigation. Binding free energy calculations revealed that 5-benzylacyclouridine exhibited the strongest binding affinity, suggesting a relatively higher predicted binding affinity (Table 3). Molecular docking visualizations for all three compounds indicated favorable predicted binding conformations with UPP1, consistent with computational docking predictions rather than experimental validation (Figure 5D–F). Additionally, CTD analysis indicated that all six key genes were strongly associated with various disease phenotypes, including delayed effects of prenatal exposure, weight loss, hepatomegaly, and inflammation (Figure 5G–L).
RT-qPCR validation of key diagnostic genes in clinical samples
To validate the expression levels of key genes, blood samples were collected from eight COPD patients and eight control subjects, and this analysis was considered a preliminary validation due to the limited sample size. As shown in Figure 6A–F, PTRF, TOR3A, and FAM168B were significantly downregulated, whereas PRKCDBP and UPP1 were significantly upregulated in COPD samples, consistent with the trends observed in the bioinformatics analysis. In contrast, no significant difference was observed for B4GALT2 expression between the two groups. This discrepancy may be attributed to the limited sample size or differences in sample types between datasets and clinical specimens.
DATA AVAILABILITY STATEMENT:
All RNA-sequencing data were obtained from the Gene Expression Omnibus database (GEO, https://www.ncbi.nlm.nih.gov), with GSE54837 selected as the training set and GSE112811 as the validation set. The code used in this analysis can be obtained from https://doi.org/10.5281/zenodo.21771476.

Figure 1: Identification of overlapping genes, enrichment analysis, and PPI network construction. (A) Volcano plot of DEGs in the GSE54837 dataset. (B) Soft threshold screening. (C) Module clustering dendrogram (after merging). (D) Heat map of the correlation between modules and traits. (E) Venn diagram for identifying overlapping genes. (F) Mulberry diagram of GO enrichment analysis, showing the main enrichment results of the intersection genes in MF, CC, and BP. (G) Lollipop diagram of KEGG signaling pathway enrichment analysis, the bubble size represents the number of enriched genes. (H) PPI network of the overlapping genes; nodes represent proteins, and edges represent protein-protein interactions. Abbreviations: DEGs = differentially expressed genes; PPI = protein-protein interaction; GO = Gene Ontology; MF = molecular function; CC = cellular component; BP = biological process; KEGG = Kyoto Encyclopedia of Genes and Genomes; ac4C-RGs = N4-acetylcytidine-related genes. Please click here to view a larger version of this figure.

Figure 2: Identification of six key genes in COPD. (A) LASSO cross-validation curve. (B) LASSO regression coefficient path diagram. As λ increases, the coefficients of unimportant genes converge to 0. (C) XGBoost feature importance ranking. The x-axis represents the gain value, the y-axis represents the gene name, and the color depth represents the importance. (D) RF feature importance ranking. The x-axis represents mean decrease in Gini. (E) The Venn diagram of overlapping genes obtained by cross-analysis of the three algorithms. Abbreviations: LASSO = least absolute shrinkage and selection operator; XGBoost = extreme gradient boosting; RF = random forest. Please click here to view a larger version of this figure.

Figure 3: Construction of a diagnostic model and expression analysis of key genes. (A) ROC curve of the training set. (B) ROC curve of the internal test set. (C) ROC curve of the external validation set. (D) Calibration curve of the training set. (E) DCA of the training set. (F) Calibration curve of the external validation set. (G) DCA of the external validation set. (H) Nomogram of six key genes. For prediction of individual COPD risk, each gene is assigned a corresponding score. (I) Expression analysis of six key genes in COPD and control samples of the GSE54837 dataset. Abbreviations: ROC = receiver operating characteristic; AUC = area under the curve; DCA = decision curve analysis; COPD = chronic obstructive pulmonary disease. Please click here to view a larger version of this figure.

Figure 4: Regulatory network and functional significance of key genes in COPD. (A) GeneMANIA analysis results of 6 key genes. The color of the lines indicates the correlation between genes, and the color of the nodes indicates different functional categories. (B) Venn diagram of cross-analysis of DIANA-microT and miRanda databases. (C) Mulberry diagram of the ceRNA regulatory network. (D) Potential transcription factor regulatory network. Blue nodes represent transcription factors, and orange nodes represent target genes. (E) Single gene GSEA enrichment analysis of UPP1, including GO. (F) Single gene GSEA enrichment analysis of UPP1, including KEGG. Abbreviations: GO = Gene Ontology; KEGG = Kyoto Encyclopedia of Genes and Genomes. Please click here to view a larger version of this figure.

Figure 5: Immune infiltration of key genes and prediction of drug targets in COPD. (A) Differences in immune cell abundance between groups. (B) Heat map of the correlation between immune cells and key genes. (C) Interaction network between key genes and predicted drugs. (D) Molecular docking of fluorouracil with UPP1. (E) Molecular docking of capecitabine with UPP1. (F) Molecular docking of 5-benzylacyclouridine with UPP1. For each compound, the left image shows the overall docking conformation, and the right image shows the local binding interactions. (G) CTD analysis of B4GALT2. (H) CTD analysis of FAM168B. (I) CTD analysis of PRKCDBP. (J) CTD analysis of PTRF. (K) CTD analysis of TOR3A. (L) CTD analysis of UPP1. Please click here to view a larger version of this figure.

Figure 6: RT-qPCR validation of key gene expression in COPD and control samples. (A) Relative expression of PTRF. (B) Relative expression of PRKCDBP. (C) Relative expression of UPP1. (D) Relative expression of TOR3A. (E) Relative expression of FAM168B. (F) Relative expression of B4GALT2. ns = not significant, p > 0.05; * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001. Abbreviation: RT-qPCR = reverse transcription quantitative polymerase chain reaction. Please click here to view a larger version of this figure.
Supplementary Figure 1: GSE54837 dataset samples and gene module clustering. (A) Sample clustering diagram of the GSE54837 dataset. (B) Module clustering dendrogram before merging. Genes were grouped using the dynamic tree cut method to identify distinct modules. (C) Hierarchical clustering dendrogram of module eigengenes. Modules with similar expression patterns were clustered based on their eigengene similarity.Please click here to download this file.
Supplementary Figure 2: GSEA enrichment analysis. (A) GO analysis of PRKCDBP. (B) KEGG analysis of PRKCDBP. (C) GO analysis of PTRF. (D) KEGG analysis of PTRF. (E) GO analysis of TOR3A. (F) KEGG analysis of TOR3A. (G) GO analysis of FAM168B. (H) KEGG analysis of FAM168B. (I) GO analysis of B4GALT2. (J) KEGG analysis of B4GALT2. Abbreviations: GO = Gene Ontology; KEGG = Kyoto Encyclopedia of Genes and Genomes.Please click here to download this file.
| Dataset | Controls | Patients | Sequencing platform |
| GSE54837 | 90 | 136 | GPL570 |
| GSE112811 | 44 | 20 | GPL570 |
Table 1: Gene expression datasets used in the study. Characteristics of the GSE54837 and GSE112811 datasets used for model development/internal testing and external validation, respectively.
| Patient | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
| Gender (F/M) | M | M | M | F | M | M | M | M |
| Age (years) | 69 | 72 | 75 | 73 | 68 | 70 | 69 | 71 |
| Smoking status | Yes | Yes | Quit smoking(2 years) | No | Yes | Yes | Yes | Quit smoking(5 years) |
| Pack-years | 20 per day / 30 years | 15 per day / 35 years | 20 per day / 50 years | | 20 per day /40 years | 30 per day / 40 years | 15 per day / 40 years | 20 per day / 30 years |
| COPD group | 2 | 3 | 3 | 2 | 2 | 3 | 2 | 3 |
Table 2: Baseline characteristics of the study participants. Baseline demographic and clinical characteristics of the COPD patients and healthy controls included in the RT-qPCR validation.
| Molecular name | Gene | Score(kcal/mol) |
| 5-Benzylacyclouridine | UPP1 | -9.6 |
| Capecitabine | UPP1 | -6.1 |
| Fluorouracil | UPP1 | -5.5 |
Table 3: Molecular docking results for UPP1 and candidate compounds.
Predicted molecular docking results for the interaction of UPP1 with fluorouracil, capecitabine, and 5-benzylacyclouridine, including their binding affinities.