Data on AMI-related gene expression matrices were searched by gene expression omnibus (GEO). We selected patients with MI for 6 months. Specifically, GSE62646 dataset was based on GPL6244 platform, containing blood samples from 28 AMI patients (6 months) and 14 patients with stable coronary artery disease without a history of myocardial infarction (MI). On the other hand, GSE59867 dataset contained blood samples from 83 AMI patients (6 months) and 46 patients with stable coronary artery disease without a history of MI, and the sequencing platform was GPL6244. In the GSE62646 and GSE59867 datasets, blood samples collected on non-first-day-of-admission were excluded to ensure that the selected disease samples were not influenced by treatment. We considered patients with stable coronary artery disease without a history of MI as the control group for this study. Meanwhile, the single cell dataset GSE269269 included peripheral blood mononuclear cells (PBMC) of 5 AMI patients without plaque rupture. Mitochondrial dynamics, which encompasses processes like mitochondrial fission, fusion, and autophagy, is essential for crucial role in cellular energy metabolism and maintenance of homeostasis. To comprehensively analyze these processes, 15 genes closely related to mitochondrial fission and 9 genes associated with mitochondrial fusion were screened using the MitoCarta 3.0 database. Additionally, 28 genes related to mitophagy were retrieved from the Reactome database. After merging and eliminating duplicates, a final set of 50 MD-RGs was identified (Supplementary Table S1). These genes constitute a valuable resource for in-depth investigations into the roles of mitochondrial dynamics in cellular physiology and pathology.

Based on a pre-screened set of (MD-RGs), we employed the single-sample gene set enrichment analysis (ssGSEA) method from the GSVA package (v 1.42.0) (13) to calculate the enrichment scores of MD-RGs for each AMI sample and control sample in the GSE62646 dataset, aiming to reflect the overall activity level of this gene set in individual samples and to compare the differences in scores between the two groups. To further explore the heterogeneity of mitochondrial dynamics states among AMI samples, we stratified all AMI samples into a high-score group and a low-score group based on the median value of their MD-RGs scores (14). Subsequent differential expression analysis was undertaken via limma package (v 3.54.1) with the thresholds of P < 0.05 and |log2 Fold Change (FC)| > 0.5, aiming to identify differentially expressed genes (DEGs) between AMI and control samples as well as between high- and low-scoring groups (15). Volcano maps and heat maps were created with the ggplot2-package (v 3.3.6) (16) and heatmap-package (v 1.0.12) (17), respectively, to demonstrate DEGs expression patterns. Next, overlapping analysis of the two sets of DEGs was completed to determine the common DEGs.

To reveal the biological functions that common DEGs were involved in the development of AMI, Gene Ontology (GO) annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were undertaken via cluster Profiler-package (v 4.6.2) (18) and the org.Hs.eg.db-package (v 3.16.0) (19) (P < 0.05). In addition, a PPI network between common DEGs was created through STRING (confidence level >0.15) to elucidate the association between them at the protein level.

Machine learning, a branch of artificial intelligence, focuses on empowering computer systems to learn from data and improve performance through the use of statistical techniques, with a wide and diverse range of applications in the biomedical field (20). In this study, we utilized two machine learning algorithms to identify feature genes closely associated with AMI from common DEGs in GSE62646 dataset. At first, the glmnet-package (v 4.1-4) (21) was utilized to perform least absolute shrinkage and selection operator (LASSO) analysis, with the parameter “family” set to binomial and “lambda” set to 0. This approach was considered as the optimal method for selecting feature genes. Secondly, Boruta algorithm, implemented through the Boruta-package (v 8.0.0), was based on the idea of random forests, where the importance of each feature was evaluated by comparing it with randomly generated “shadow” features (22). During the screening process using the Boruta algorithm, the program defined DEGs, categorizing them into accepted and rejected groups. The accepted genes represented the feature genes selected by the Boruta model. After that, the intersection of two sets of feature genes was extracted with the help of Venndiagram-package (v 1.7.3), which were recorded as candidate genes (23).

In GSE62646 and GSE59867 datasets, the expression trends of candidate genes were further compared between AMI and control samples utilizing the Wilcoxon test. Our focus was on genes exhibiting stable expression, characterized by consistent trends across both datasets and significant group differences (P < 0.05). Further, to assess the diagnostic accuracy of the model, ROC curves were generated based on gene expression in both datasets with the use of the ROC-package (v 1.18.0), and genes with the area under curve (AUC) greater than 0.7 were defined as biomarkers associated with mitochondrial dynamics in AMI (24). In general, the AUC greater than 0.7 indicated that the gene was more effective in diagnosing the disease. Subsequently, the biomarkers were input into Gene MANIA online website to generate a network, aiming to explore potential mechanisms of biomarkers in AMI.

To predict the occurrence of AMI from perspective of the biomarkers as a whole, we integrated the expression of the biomarkers and subsequently constructed a nomogram in GSE62646 dataset applying the rms-package (v 6.5-0) (25). Furthermore, calibration curve and ROC curve were created to estimate the accuracy of nomogram predictions, as well as decision curve analysis (DCA) was completed to determine the likelihood of clinical benefit from the nomogram.

GSEA was conducted to identify gene sets associated with specific biological processes, pathways, or functions related to biomarkers. The KEGG gene set (c2.cp.kegg.v7.4.symbols.gmt) retrieved from MsigDB database served as the background gene set, and gene set enrichment analysis (GSEA) was implemented with cluster Profiler-package. Briefly, based on all samples from GSE62646, Spearman correlation coefficients between each biomarker and the remaining genes were computed by the psych-package (v 2.1.6), followed by ranking the genes according to the coefficients (from highest to lowest). Then, the ranked genes were treated as the gene set to be detected, and the GSEA was then completed. The gene set with P < 0.05 and q < 0.25 was considered significant.

To explore the immune profile in AMI, CIBERSORT method was employed to infer the relative proportions of 22 immune-infiltrating cells in each sample for GSE62646 dataset (26, 27). Next, differences in proportions of immune-infiltrating cells between AMI and control groups were compared by Wilcoxon test (P < 0.05), followed by drawing box plot using ggplot2-package to visualize the results. Following this, Spearman correlation analysis was completed between the differential immune cells as well as between the differential immune cells and biomarkers to explore their associations, with thresholds set at |cor| > 0.3 and P < 0.05.

To elucidate the molecular regulatory mechanisms of the biomarkers, microRNAs (miRNAs) regulating the biomarkers were predicted by applying the miRDB and TargetScan databases. Next, starBase and miRNet databases were utilized to retrieve lncRNAs that bound to the miRNAs obtained as described above. In addition, transcription factors (TFs) targeting biomarkers were collected through accessing University of California, Santa Cruz Genome Browser (UCSC). By integrating the relationship pairs obtained above, the regulatory networks of lncRNA-miRNA-mRNA and TF-biomarker were created via Cytoscape software (v 3.9.0) (28).

The m6A regulators are crucial for post-transcriptional gene expression regulation, which in turn affects a series of physiological and pathological functions. Related research has emphasized that m6A can participate in regulating the development of AMI (29). Therefore, this study used the Wilcoxon test to compare expression differences of 13 widely reported m6A regulators between AMI and control groups in GSE62646 dataset (P < 0.05). The 13 m6A regulators were METTL3, METTL14, WTAP, KIAA1429, RAMI15, ZC3H13, YTHDC1, YTHDC2, YTHDF1, YTHDF2, HNRNPC, FTO, and ALKBH5. Thereafter, Spearman correlation analysis was carried out for biomarkers and m6A regulators in AMI samples using the psych-package (P < 0.05).

AMI is the most common public health cardiovascular disease with a high number of complications. Therefore, diseases significantly associated with biomarkers (top 20) were analyzed using comparative toxicogenomics database (CTD), and the results were visualized by Cytoscape. In addition, small molecule drugs targeting biomarkers were predicted by thorough analysis of the DGIdb database. Subsequent molecular docking was accomplished to investigate the feasibility of treating AMI. Specifically, it was possible to obtain the 3D structures of small-molecule medications applying PubChem database. Next, Protein Data Bank (PDB) database was applied to acquire the protein crystal structures that corresponded to the biomarkers. Ultimately, molecular docking procedure was executed by applying CB-Dock2, and docking binding energy was calculated. Docking results were generally considered feasible for docking binding energies less than −5 kcal/mol.

The data of AMI samples from the GSE269269 dataset were filtered for scRNA-seq analysis by the “Seurat” package (v 5.1.0) (30). Specifically, the criteria for filtered were that (1) low quality cells with genes less than 200 and abnormally highly expressed cells with genes greater than 5,000; (2) the sum of the expression of all genes gauged for per cell was ≥25,000 or <500; (3) cells with percentage of mitochondria greater than 25%. In order to obtain highly variable genes (HVGs), the above filtered data were normalized by the NormalizeData function in “Seurat” package (v 5.1.0) to obtain the standardized data. Then, the FindVariableFeatures function was carried out to select the top 2,000 HVGs. The result was visualized and the top 10 genes with highest intercellular expression variation were labeled by the LablePoints function. To identify the statistically significant principal components, data scaling was carried out with ScaleData function in “Seurat” package (v 5.1.0). After that, a principal component analysis (PCA) plot was drawn between all the samples to confirm whether the overall distribution of cells was consistent per sample and whether significant outlier samples existed. At the same time, PCA was also conducted using these 2,000 HVGs. Then, JackStrawPlot function served to contrast the p-values distribution of each principal components (PCs) (p < 0.05). The scree plot was visualized by applying the ElbowPlot function, and the number of PCs corresponding to where the curve began to level off were selected for uniform manifold approximation and projection (UMAP) cell cluster analysis (resolution = 0.8). Furthermore, use the FindAllMarkers function in “Seurat” package (v 5.1.0) to identify the marker genes in each cell cluster of the single—cell dataset (logfc.threshold = 0.5, min.pct = 0.25, return.thresh = 0.01). Then, annotating different cell clusters to different cell types based on the marker genes using the “singleR” package (v 2.4.0) (31) and the CellMarker database. The bubble plots depicting different marker gene expression in different cell types were drawn. Finally, to identify key cells associated with key genes, UMAP plots, bubble plots and scatter plots were created to show biomarkers expression on all cell types in AMI samples. The cells in which the expression levels of biomarkers were relatively high and had large cell number of cells were defined as key cells.

Cell communication analysis was conducted to examine the expression and pairing of receptors and ligands in different cells, aiming to infer interactions between different cells. The communication network between cells were analyzed by the “CellChat” package (v 1.6.1) (32). To gain deeper insights into heterogeneity of the key cells, based on key cells, in GSE269269, the key cells were regrouped, the cells were annotated into different cell subsets, and the clustering results were visualized using UMAP. In order to investigate the differentiation trajectory and evolution of key cells during development, the “monocle” package (v 2.26.0) (33) was utilized for pseudo-time analysis to investigate alterations in the developmental trajectory of key cells and variations in the expression of biomarkers.

RT-qPCR was performed to detect and analyze the expression levels of specific mRNAs in blood samples.The study was conducted in accordance with the Declaration of Helsinki, and RNA was collected from 5 pairs of blood samples, including 5 acute myocardial infarction samples and 5 control samples. Consistent with the previous data, we selected patients with myocardial infarction for 6 months. According to the corresponding age and gender, healthy people were selected as controls. The agency responsible for ethical approval was designated as Qilu hospital. The participants all completed and signed an informed consent form, and the ethical approval agency was Ethics Committee of Qilu Hospital of Shandong University [approval no. KYLL-2022(ZM)-082]. Since patient genetic data are involved, we kept patient-specific information confidential. Whole blood samples were collected from patients using EDTA-anticoagulant tubes. Total RNA was extracted from whole blood samples using the RNAprep Pure Hi-Blood Kit (DP443, Tiangen Biotech, Beijing, China). Briefly, erythrocytes were lysed, and leukocytes were homogenized in a proprietary lysis buffer. After ethanol addition, the lysate was applied to a silica-membrane column for RNA binding. The column was washed, treated with DNase I to remove genomic DNA contamination, washed again, and total RNA was finally eluted in RNase-free water. RNA concentrations were measured by NanoPhotometer N50. Secondly, mRNA was transcribed to synthesize cDNA using SweScript First Strand cDNA synthesis kit (Servicebio, Wuhan, China). For Quantitative RT-PCR analysis, each reaction mixture consisted of 3 µL cDNA, 5 µL 2x Universal Blue SYBR Green qPCR Master Mix (Servicebio), 1 µL forward primer (10 µM), and 1 µL reverse primer (10 µM) (Supplementary Table S2). The mRNA levels were normalized by GAPDH which was selected as an endogenous reference gene. The relative gene expression was calculated by 2-^△△Ct, and p-value was calculated by Graphpad Prism 5.

Group comparisons were assessed using the Wilcoxon test. Statistical discrepancies were recognized as P-value below 0.05. In this work, bioinformatic statistical analyses were implemented employing R program (version 4.2.3).

Within the GSE62646 dataset, differential analysis revealed a significant decrease in MD-RGs scores in AMI samples compared to control samples (p < 0.001) (Figure 1A). This might suggest that the expression of MD-RGs may influence the progression of disease in patients with AMI. In the AMI sample for the GSE62646 dataset, there was a marked discrepancy in MD-RGs scores between high- and low-scoring groups divided based on the ssGSEAs algorithm (P < 0.05) (Figure 1B). By setting thresholds of P < 0.05 and |log2FC| > 0.5, 55 DEGs (43 up-regulated and 12 down-regulated in high-scoring groups) between high- and low-scoring groups (Figures 1C,D), as well as 157 DEGs (118 up-regulated and 39 down-regulated in AMI samples) were mined between AMI and control samples in GSE62646 dataset (Figures 1E). Following this, the Venn diagram displayed 29 common DEGs by overlapping the two sets of DEGs (Figure 1F). These 29 genes likely represented a gene set that played roles in both the occurrence of AMI and different mitochondrial functional states.

These 29 common DEGs were enriched and analyzed, yielding 9 GO entries and 13 KEGG signaling pathways (P < 0.05). With respect to GO, the entries were mainly related to “cellular detoxification”, “detoxification”, “response to toxic substance”, “cellular oxidant detoxification”, “hydrogen peroxide catabolic process”, “peroxidase activity”, and so on (Figure 2A, Supplementary Table S3). KEGG analysis elucidated that candidate genes were engaged in “Chemical carcinogenesis—reactive oxygen species”, “Glutathione metabolism”, “Drug metabolism—cytochrome P450”, etc. (Figure 2B, Supplementary Table S4). In addition, the PPI network of the 29 common DEGs, containing 15 points and 17 edges, was illustrated in Figure 2C, and it could be noted that H4C6, PRDX1, and NAMPT seemed to have stronger interactions with the remaining genes. This suggested that these genes might occupy pivotal positions in co-regulated biological processes and merited further investigation.

These 29 common DEGs were incorporated into two machine learning algorithms to identify feature genes. The LASSO algorithm identified nine feature genes associated with AMI at lambda. min = 0.000299114, namely SNORA24, SNORD15B, CLC, RN7SL368P, SNORD54, COX7B, NAMPT, MGST3, and CCDC97 (Figure 3A). Furthermore, Boruta algorithm screened out 19 feature genes based on each feature's importance, as shown in Figure 3B. We utilized a Venn diagram to identify common genes in the two algorithms, yielding a total of eight genes (SNORA24, SNORD15B, RN7SL368P, COX7B, NAMPT, MGST3, CCDC97, and SNORD54), which were recorded as candidate genes for further exploration (Figure 3C). Given this, the integration of two machine learning algorithms effectively narrowed down the range of key genes, thereby enhancing the focus of subsequent analyses.

The expression analyses of candidate genes were conducted in GSE62646 and GSE59867 datasets, which showed that six genes were markedly distinct between AMI and control groups (P < 0.05) (Figures 3D,E). Among them, CCDC97 was up-regulated in AMI samples, while the remaining five genes (COX7B, NAMPT, RN7SL368P, SNORD54, and SNORD15B) were down-regulated in AMI samples. What's more, the AUC values of two (COX7B and SNORD54) of these six genes in both datasets were greater than 0.7 through the ROC curves (Figures 3F,G), implying that, they could better differentiate between AMI and control samples. Therefore, COX7B and SNORD54 were considered as biomarkers associated with mitochondrial dynamics in AMI. In addition, an interaction network between COX7B and 20 highly related genes was generated on the Gene MANIA platform, with 8.01% co-expression, 77.64% physical interactions, 5.37% prediction, 3.63% co-localization, 2.87% genetic interactions, 1.88% pathways, and 0.60% shared protein structural (Figure 3H). Unfortunately, the interaction network of SNORD54 was not predicted. Finally, we estimated the expression of 2 biomarkers by RT-qPCR. The expression levels of COX7B and SNORD54 were significantly downregulated in peripheral blood mononuclear cells (PBMCs) of AMI samples compared to control (P < 0.05), consistent with its expression in GSE62646 and GSE59867 (Figure 3I). In summary, both COX7B and SNORD54 demonstrated stable differential expression at the transcriptional level and in clinical samples, along with good diagnostic differentiation capability, establishing a foundation for their potential as biomarkers for AMI.

We created a nomogram model in GSE62646 dataset so as to facilitate the clinical prediction of AMI using the selected two biomarkers (Figure 4A). In the nomogram, the lower the expression of the two biomarkers and the higher the total score, the higher the risk of developing AMI. The slope of the calibration curve was close to 1 and AUC value of ROC curve was 0.99, both of which indicated that the accuracy of the nomogram in predicting AMI was superior (Figures 4B,C). The DCA results revealed that the nomogram yielded a net benefit superior to that of the individual biomarker (Figure 4D). In conclusion, nomogram integrated by two biomarkers exhibited high potential clinical value.

In order to elucidate the biological mechanisms of two biomarkers in AMI, we completed GSEA in the GSE62646 dataset. Results indicated that the expression of two biomarkers was linked to “calcium signaling pathway”, “oxidative phosphorylation”, “MAPK signaling pathway”, “Wnt signaling pathway”, “cell cycle”, “Notch signaling pathway”, “cytokine-cytokine receptor interaction”, “ECM-Receptor interaction”, and other pathways (P < 0.05 and q < 0.25) (Supplementary Table S5-S6). The top 5 pathways were selected for visualization based on the order of significance (Figures 5A,B). These pathways and biological processes were intertwined and collectively participate in the onset, progression, and subsequent pathology of AMI. A deeper understanding of the role of these pathways could provide a theoretical basis for the discovery of novel therapeutic strategies in the future.

The development of AMI might be accompanied by abnormalities in the proportion and function of immune cells, so we mined the links between two biomarkers and 22 immune cells. The infiltration content of 22 immune cells in each sample of the GSE62646 dataset was calculated by the CIBERSORT algorithm, as demonstrated in Figure 5C. After removing the cells that turned out to be 0 in 30% of the samples, the expression of the remaining 16 immune cells in AMI and control samples was compared applying the Wilcoxon test, which showed that five immune cells were markedly distinct between groups (P < 0.05) (Figure 5D). Among them, M0 macrophages, naive CD4 T cells, and regulatory T cells (Tregs) were more abundant in AMI, while monocytes and neutrophils showed the opposite trend. In addition, the strongest positive association was presented between neutrophils and monocytes (cor = 0.53), and the highest negative association was presented between monocytes and M0 macrophages (cor = −0.43) (Figure 5E). Notably, a marked positive association was observed between SNORD54 and Tregs (cor = 0.307, P < 0.05) (Figure 5F). These findings revealed a specific pattern of immune cell imbalance in AMI and preliminarily established an association between biomarkers and the immune microenvironment.

The mRNA-miRNA-lncRNA relationship network was composed of one mRNA (COX7B), five miRNAs, and 13 lncRNAs (Figure 6A). All five miRNAs (hsa-mir-558, hsa-mir-520f-3p, hsa-mir-556-5p, hsa-mir-607, hsa-mir-192-3p) targeted COX7B; unfortunately, no miRNAs regulating SNORD54 were predicted. In addition, it could be found that lncRNAs (KCNQ1OT1, EBLN3P, LINC01235, etc.) regulation of COX7B was achieved through hsa-mir-520f-3p, as well as NEAT1 could simultaneously regulate COX7B through hsa-mir-520f-3p and hsa-mir-556-5p. In total, 35 TFs were retrieved, including 21 TFs regulating COX7B and 16 TFs regulating SNORD54, and the network of TF-biomarker was presented in Figure 6B. It could be observed that two TFs (KDM5A and CEBPG) could regulate two biomarkers simultaneously. This analysis preliminarily constructed a molecular network framework suggesting that COX7B and SNORD54 may be regulated by non-coding RNAs and transcription factors, thereby laying the groundwork for subsequent mechanistic investigations.

The m6A modifications palyed a pivotal role in the regulation of gene expression after AMI and might affect inflammatory responses, apoptosis, and cardiac regenerative repair processes. In this study, six m6A regulators (HNRNPC, KIAA1429, METTL3, WTAP, YTHDC1, and YTHDC2) were found to be significantly different between AMI and control samples in GSE62646 dataset (P < 0.05) (Figure 7A), and all of them were under-expressed in AMI samples. Notably, COX7B had the highest marked positive association with WTAP (cor = 0.62, P < 0.001), and SNORD54 had the strongest significant positive association with YTHDC1 (cor = 0.69, P < 0.001) (Figure 7B). These correlations suggest that the dysregulation of COX7B and SNORD54 expression may co-vary with the extensive alterations in m6A modification observed in AMI.

A comprehensive analysis of CTD database was conducted to retrieve diseases significantly linked to the two biomarkers, which revealed that chromosome-defective, nerve degeneration, and hyperplasia could be predicted by both biomarkers (Figure 8A). Additionally, SNORD54 could be found to be significantly associated with cardiovascular diseases such as ventricular dysfunction (left), heart diseases, and myocardial infarction. Further, the prediction of the DGIdb database showed that 25 drugs related to the biomarkers were retrieved, of which the top 3 drugs significantly related to COX7B were 1-Methyl-4-phenyl-2,3-dihydropyridinium, cube root extract, 3'-Azido-3'- deoxythymidine, and the top 3 drugs significantly associated with SNORD54 were chrysin, quercetin, and harmine (Figure 8B). The molecular docking results revealed that the binding energies of COX7B with 1-Methyl-4-phenyl-2,3-dihydropyridinium, cube root extract, and 3'-Azido-3'-deoxythymidine were −8.1 kcal/mol, −7.5 kcal/mol, and −10.9kcal/mol, accordingly, implying that COX7B had a strong affinity for all three drugs (Figures 8C–E). Because SNORD54 had no protein structure, molecular docking was not performed. This section of the results provides preliminary leads for future exploration of compounds targeting COX7B and the potential roles of drugs related to SNORD54 (such as chrysin and quercetin) in AMI.

We screened single-cell sequencing data from 5 AMI samples, resulting in a total of 26,957 genes and 33,162 cells. The results before and after QC were showed in Supplementary Figures S1A, S1B. After standard data processing, we selected 2,000 HVGs for subsequently analysis, of which the top 10 genes exhibiting the most pronounced intercellular expression changes were labeled (Supplementary Figure S1C). There were no obvious outlier samples (Supplementary Figure S1D). The top 30 PCs were chosen for further analysis (Supplementary Figure S1E–F). Then, a sum of 26 different cell clusters were identified (Figure 9A), and 26 clusters were annotated to 10 cell types, namely T cells, natural killer (NK) cells, monocytes, neutrophils, B cells, immature erythroid cells, neutrophil progenitor cells, basophils, dendritic cells and mast cells (Figure 9B). The bubble plot demonstrated that the marker genes exhibited high specificity (Figure 9C). Among the two biomarkers, only COX7B existed in the single-cell dataset. Therefore, COX7B was used to search for key cells. According to the results of Figures 9D–F and Table 1, the number of monocytes is large, and the expression level of COX7B in monocytes was the highest. Meanwhile, the number of NK cells was also relatively large, and the expression of COX7B was also relatively high. Therefore, monocytes and NK cells were selected as the key cells.

We conducted an analysis of intercellular communication networks between different cell types in AMI groups. Particularly, all cell types interacted with each other. In terms of the number of interactions, mast cells had the strongest interaction with monocytes, and NK cells had relatively strong interactions with both monocytes and mast cells (Figure 10A). In terms of interaction weights, monocytes also had the strongest correlation with mast cells, and NK cells had the strongest correlation with monocytes (Figure 10B). This suggests that within the AMI microenvironment, monocytes and NK cells exhibit active crosstalk with other immune cells, particularly mast cells, potentially working together to orchestrate the immune response.

In GSE269269, secondary clustering based on monocytes (resolution = 0.3) and NK cells (resolution = 0.8) was performed, respectively, reclassifying the monocytes into 13 distinct subtypes and the NK cells into 12 distinct subtypes (Figures 11A,B). In addition, in the results of the pseudo-time trajectory inference analysis, the trajectories of monocytes were shown in Figure 11C. Cells differentiate from right to left over time. Monocytes were found to include a total of 10 different states of differentiation and 13 subtypes. For monocytes, state 1 was the early stage of its development. The COX7B gene was highly expressed throughout all states (Figure 11D). Meanwhile, the trajectories of NK cells were shown in Figure 11E. Cells differentiate from up to down over time. NK cells were found to include a total of 9 different states of differentiation and 12 subtypes. For NK cells, state 1 was the early stage of its development. The COX7B gene was also highly expressed throughout all states (Figure 11F). Pseudotime analysis indicated that COX7B maintained consistently high expression across different differentiation stages and subtypes of monocytes and NK cells, suggesting that its role in the basic functions of these immune cells may be important in the context of AMI.