Identification of common mechanisms and biomarkers of atrial fibrillation and heart failure based on machine learning

Abstract

Aims

Atrial fibrillation (AF) is the most common arrhythmia. Heart failure (HF) is a disease caused by heart dysfunction. The prevalence of AF and HF were progressively increasing over time. The co‐existence of AF and HF presents a significant therapeutic challenge. In order to provide new ideas for the diagnosis of AF and HF, it is necessary to carry out biomarker related studies.

Methods and results

The training set and validation set data of AF and HF patient samples were downloaded from the GEO database, ‘limma’ was used to compare the differences in gene expression levels between the disease group and the normal group to screen for differentially expressed genes (DEGs). Weighted correlation network analysis (WGCNA) identified the modules with the highest positive correlation with AF and HF. Functional enrichment and PPI network construction of key genes were carried out. Biomarkers were screened by machine learning. The infiltration of immune cells in AF and HF groups was evaluated by R‐packet ‘CIBERSORT’. The miRNA network was constructed and potential therapeutic agents for biomarker genes were predicted through the drugbank database. Through WGCNA analysis, it was found that the modules most positively correlated with AF and HF were MEturquoise (r = 0.21, P value = 0.09) and MEbrown (r = 0.62, P value = 8e‐12), respectively. We screened 25 genes that were highly correlated with both AF and HF. Lasso regression analysis results showed 7 and 20 core genes in AF and HF groups, respectively. The top 20 important genes in AF and HF groups were obtained as core genes by RF model analysis. Four biomarkers were obtained after the intersection of core genes in four groups, namely, GLUL, NCF2, S100A12, and SRGN. The diagnostic efficacy of four genes in AF validation sets was good (AUC: GLUL 0.76, NCF2 0.64, S100A12 0.68, and SRGN 0.76), as well as in the HF validation set (AUC: GLUL 0.76, NCF2 0.84, S100A12 0.92, and SRGN 0.68). The highest correlation with neutrophils was observed for GLUL, NCF2, and S100A12, while SRGN exhibited the strongest correlation with T cells CD4 memory resting in the AF group. GLUL, NCF2, S100A12, and SRGN were most associated with neutrophils in the HF group. A total of 101 miRNAs were predicted by four genes, and GLUL, NCF2, and S100A12 predicted a total of 10 potential therapeutic agents.

Conclusions

We identified four biological markers that are highly correlated with AF and HF, namely, GLUL, NCF2, S100A12, and SRGN. Our findings provide theoretical basis for the clinical diagnosis and treatment of AF and HF.

Introduction

Atrial fibrillation (AF) is a heart disease that causes irregular and abnormally rapid heart rates.¹ It is a common type of tachyarrhythmia. AF increases the risk of stroke, heart failure, and other complications related to the heart.² Heart failure (HF) is a group of syndromes caused by ventricular dysfunction,³ which results from heart malfunction and is considered a modern epidemic with high morbidity and mortality rates among patients.⁴ AF and HF frequently co-exist, posing a challenge for effective treatment. A comprehensive and in-depth comprehension of the underlying pathogenesis of AF and HF is imperative for optimizing and advancing novel therapeutic approaches.

Epidemiological studies have found that the main risk factors for AF include aging, a sedentary lifestyle, obesity, diabetes, metabolic syndrome, and obstructive sleep apnoea.^5,6 However, more and more studies have found that genetics may be a risk factor for susceptibility to AF. AF typically exhibits a strong familial inheritance and a high clinical incidence. For example, Lubitz et al. found that AF has a certain heritability, and the incidence of familial AF patients is higher than that of non-familial AF patients.⁷ Ragab et al. reported an increased risk of developing AF in individuals with a parental history of AF.⁸ In addition, based on molecular studies, it was found that the abnormal expression of PI3Ks, SLC7A11, CXCL12, and other genes participated in the pathogenesis of AF.^9–11 Therefore, understanding the regulatory genes of AF can aid in early identification of individuals at high risk for developing AF. HF is a cardiovascular disease characterized by symptoms such as fatigue, significantly reduced exercise endurance, oedema, and paroxysmal nocturnal dyspnoea.¹² The morbidity and mortality of HF-related diseases are currently increasing globally, with HF being the leading cause of death. Rosik et al. identified SERCA2a, S100A1, and IPP-1 as potential therapeutic targets for HF.¹³ Bioinformatics studies have identified NSG1, PHLDA1, and SERPINE2 as potential biomarkers of HF.¹⁴ Previous studies have shown that AF is the most common arrhythmia in patients with HF, and HF is the most common clinical cause of death in patients with AF. It is crucial to screen common biomarkers of AF and HF for disease diagnosis. The research found that LASSO and RF algorithms play a crucial role in handling binary classification problems. In addition, studies have reported that LASSO and RF algorithms play an indispensable role in the process of screening disease biomarkers. For example, Li et al. obtained eight diagnostic genes for Alzheimer's disease and metabolic syndrome through RF and Lasso learning methods.¹⁵ Based on RF and LASSO algorithms, Li et al. found four biomarkers associated with HF NETs (CXCR2, FCGR3B, VNN3, and FPR2).¹⁶ However, the potential pathogenesis and biomarkers of AF and HF have not been fully elucidated, and there are still many related genes to be identified, so it is necessary to find biomarkers with good specificity and sensitivity.

In this study, the potential relationship between AF and HF was analysed using bioinformatics methods. The biomarkers GLUL, NCF2, S100A12, and SRGN were identified as highly correlated with both AF and HF. Furthermore, their clinical prognostic value was verified. These study results provide a theoretical basis for the clinical diagnosis and treatment of AF and HF.

Materials and methods

Data source

Training set: AF gene expression data from the GSE41177 and GSE79768 datasets were downloaded from the GEO database (). The GSE41177 dataset included 16 AF patient samples and 3 patients with normal sinus heart rate, and the GSE79768 dataset included 7 AF patient samples and 6 patients with normal sinus heart rate. Transcriptome data of the GSE66360 dataset for HF patients were downloaded from the GEO database, including 49 HF patient samples and 50 control samples.

Validation set: The GSE14975 dataset was downloaded from the GEO database and included 5 AF patient samples and 5 control samples. GSE175764 is a high-throughput sequencing data from Illumina NextSeq 500, consisting of 5 HF patient samples and 5 control samples.

Screening of differentially expressed genes

After the original data from GSE41177 and GSE79768 datasets were de-batch processed by R software ‘affy’, ‘limma’ package was used to compare the differences in gene expression levels between the AF group and the normal group. In the GSE66360 dataset, the ‘limma’ package was used to screen DEGs of HF group and normal group. The difference screening criteria were |log2FC| > 1, adj. P value < 0.05. Volcano maps and heat maps were drawn with ‘ggplot2’ and ‘pheatmap’, respectively, to show the expression of DEGs.

Screening of common differentially expressed genes in atrial fibrillation and heart failure

WGCNA can be effectively employed for the identification of gene sets exhibiting strong covariation. The WGCNA networks of GSE41177, GSE79768, and GSE59867 were constructed to identify the co-expression modules and genes associated with AF and HF, respectively. DEGs1 was obtained by the intersection of DEGs and genes in the module with the highest positive correlation in AF samples. DEGs2 was obtained by the intersection of DEGs in HF and the genes within the module exhibiting the highest positive correlation with HF. The intersection of DEGs1 and DEGs2 was used to obtain DEGs3 of AF and HF. The intersection result was displayed using a Venn diagram.

Functional enrichment and protein-protein interaction network construction of differentially expressed genes 3

GO and KEGG enrichment analysis of DEGs3 was performed through the ‘clusterProfile’ package. Both GO and KEGG annotated datasets are from MSigDB (v2023.1), the enrichment results were screened with P value < 0.05 as the threshold value. The results of enrichment analysis were visualized. The STRING database can explore the protein interaction network and reveal the complex molecular regulation network in the cell. The PPI network of DEGs3 was constructed by using the online website of STRING () to further understand the interaction relationship between common differential genes (maximum number of interactors = 0 and confidence score ≥ 0.4). The molecular interaction network was visualized via Cytoscape software (version 3.9.1).

Biomarkers were screened by Lasso and random forest algorithms

In order to screen important diagnostic biomarkers related to AF and HF, R packages ‘glmnet’ and ‘randomForest’ were used to perform two kinds of machine learning for Lassso regression and RF for DEGs, respectively. Both LASSO and RF analyses were conducted using 10-fold cross-validation. After LASSO analysis, genes with non-zero regression shrinkage coefficients were selected, and after importance ranking in RF analysis, the top 20 genes were selected as important genes for this algorithm. The intersection genes obtained by the two algorithms can be used as biomarker genes. ROC validation of the biomarker genes was also performed in the GSE14975 and GSE175764 validation sets.

Immunoinfiltration assay

CIBERSORT is used to estimate the relative abundance of different immune cell types from gene expression data. In the AF and HF groups, immune cell infiltration was assessed using the R-packet ‘CIBERSORT’ calculation method, and Pearson correlation analysis was used to evaluate the relationship between different immune cell phenotypes and biomarkers.

microRNA network construction

miRcode provides ‘whole transcriptome’ human microRNA target prediction based on comprehensive GENCODE gene annotation. The miRNAs associated with the biomarker genes in AF and HF were obtained through the miRcode database. The miRNA networks of the genes were visualized using Cytoscape software.

Drug prediction

The DrugBank database contains available FDA-approved drugs and experimental compounds. It provides detailed drug and drug target information. Potential therapeutic drugs were predicted by searching for biomarker genes in DrugBank's database.

Statistical analysis

R (, 4.0.2 version) was used for all statistical analyses. The data from different groups were compared by Wilcoxon test, and P < 0.05 was considered as statistically significant.

Result

Screening of differentially expressed genes in atrial fibrillation and heart failure

As shown in Figure 1A,B, there were a total of 505 DEGs in AF, including 469 up-regulated genes and 36 down-regulated genes. The differences in gene expression levels between the HF group and the normal group were compared in GSE66360 dataset, and the results were shown in Figure 1C,D. The results showed that there were 755 DEGs, 538 up-regulated genes, and 217 down-regulated genes in the HF samples. ‘ggplot2’ and ‘pheatmap’ were used to draw volcano maps and heat maps, respectively, to show the expression of DEGs.

[IMAGE OMITTED. SEE PDF]

Identification of common differentially expressed genes in atrial fibrillation and heart failure by weighted gene co-expression network analysis

The weighted gene co-expression network of GSE41177 and GSE79768 combined datasets, and GSE59867 datasets were constructed to identify co-expression modules and genes associated with AF and HF. As shown in Figure 2A–F, the highest positive correlation with AF was MEturquoise (r = 0.21, P value = 0.09), and the module with the highest positive correlation of HF was MEbrown (r = 0.62, P value = 8e-12).DEGs1 as obtained by intersections of DEGs and genes of MEturquoise module in AF, a total of 475 genes were identified, as shown in Figure 2G. According to the analysis of Figure 2H, a total of 133 DEGs2 were obtained by the intersection of DEGs and genes of MEbrown module in HF. After the intersection of DEGs1 and DEGs2, 25 common DEGs of AF and HF were obtained, which were named DEGs3, including FCER1G, LYZ, GLUL, DUSP1, CST3, SRGN, SERPINA1, S100A8, S100A9, FCGR2A, CYBB, FCGR3B, TYROBP, CXCL1, FCN1, BCL2A1, S100A12, VNN2, IGSF6, NCF2, FCGR2C, PILRA, TLR4, MPEG1, and CMTM2, as shown in the Venn diagram in Figure 2I.

[IMAGE OMITTED. SEE PDF]

Functional enrichment analysis and protein-protein interaction network network construction of differentially expressed genes 3

As shown in Figure 3A,B, a total of 371 pathways were enriched, including 291 BP pathways, 28 CC pathways, 46 MF pathways, and 6 KEGG pathways. DEGs3 were enriched in IGG_BINDING, IMMUNOGLOBULIN_BINDING, TOLL_LIKE_RECEPTOR_4_BINDING, NEUTROPHIL_CHEMOTAXIS, LEUKOCYTE_CHEMOTAXIS, MYELOID_LEUKOCYTE_MIGRATION and NEUTROPHIL_MIGRATION,ect significantly. Next, Figure 3C shows the PPI network of DEGs3.

[IMAGE OMITTED. SEE PDF]

Machine learning screening for biomarkers

Two machine learning algorithms of Lasso regression and RF were used to analyse 25 DEGs using R packages ‘glmnet’ and ‘randomForest’. Seven and 20 key genes were obtained in AF and HF groups, respectively, through Lasso regression analysis, and the results were shown in Figure 4A,B. The importance of 25 genes in AF and HF groups was ranked through RF analysis, and top 20 genes were selected as the important genes of the algorithm (Figure 4C,D). By intersecting the important genes obtained from the two models, four biomarker genes were obtained, namely, GLUL, NCF2, S100A12, and SRGN, as shown in Figure 4E. ROC validation was performed on four biomarker genes in the GSE14975 and GSE175764 validation sets. Figure 4F results showed that the AUC of GLUL, NCF2, S100A12, and SRGN in the AF group were 0.76, 0.64, 0.68, and 0.76, respectively, while that in the HF group were 0.76, 0864, 0.92, and 0.68, respectively, indicating that the four genes had good diagnostic performance in AF and HF groups.

[IMAGE OMITTED. SEE PDF]

Immunoinfiltration analysis

The infiltration of immune cells in AF and HF groups was evaluated, and it was found that the AF group had significant differences in B cells naive, B cells memory, T cells CD4 memory activated, monocytes, macrophages M1, mast cells resting, mast cells activated, and neutrophils (Figure 5A), while there were significant differences in T cells CD8, T cells CD4 memory resting, T cells follicular helper, T cells gamma delta, NK cells activated, monocytes, macrophages M2, dendritic cells activated, mast cells resting, mast cells activated, and neutrophils of the HF group (Figure 5C). The relationship between the four biomarkers and different immune cell phenotypes was analysed by Pearson. As shown in Figure 5B, the highest correlation with neutrophils was observed for GLUL, NCF2, and S100A12, while SRGN exhibited the strongest correlation with T cells CD4 memory resting in the AF group. GLUL, NCF2, S100A12, and SRGN were most associated with neutrophils in the HF group (Figure 5D).

[IMAGE OMITTED. SEE PDF]

microRNA network construction

The miRNAs related to GLUL, NCF2, S100A12, and SRGN were screened through miRcode database, and 101 miRNAs were obtained. In addition, the miRNA network of genes was visualized by Cytoscape software, as shown in Figure 6.

[IMAGE OMITTED. SEE PDF]

Drug prediction

The DrugBank database was used to predict potential therapeutic agents for biomarker genes, and the results were shown in Table 1. GLUL, NCF2, and S100A12 predicted 10 potential therapeutic agents, which were l-glutamine, glutamic acid, methionine, capsaicin, ceftriaxone, pegvisomant, ammonia, amlexanox, olopatadine, and dextromethorphan, respectively.

Table 1 Potential therapeutic drugs for biomarkers

Gene	Drugbank ID	Name	Drug group	Actions	Details	Structure
GLUL	DB00130	L-Glutamine	Approved, investigational, nutraceutical	Product	C5H10N2O3
GLUL	DB00142	Glutamic acid	Approved, nutraceutical		C5H9NO4
GLUL	DB00134	Methionine	Approved, nutraceutical	Inhibitor	C5H11NO2S
GLUL	DB06774	Capsaicin	Approved	Inducer	C18H27NO3
GLUL	DB01212	Ceftriaxone	Approved	Inducer	C18H18N8O7S3
GLUL	DB00082	Pegvisomant	Approved	Inhibitor	C990H1532N262O300S7
GLUL	DB11118	Ammonia	Approved	Substrate	\
S100A12	DB01025	Amlexanox	Approved, investigational, withdrawn	Antagonist	C16H14N2O4
S100A12	DB00768	Olopatadine	Approved	Antagonist	C21H23NO3
NCF2	DB00514	Dextromethorphan	Approved	Inhibitor	C18H25NO

Discussion

In recent years, the incidence of AF and HF has been increasing. AF can lead to HF, but it can also be a secondary complication of HF.¹⁷ AF can increase the risk of HF, and the cardiac remodelling and cardiac nerve dysfunction caused by HF can aggravate the development of AF. Therefore, patients with HF and AF exhibit more severe clinical manifestations and experience a worse prognosis. The new biomarkers revealed by multi-omics analysis can be used for the early diagnosis and drug target screening of AF and HF.¹⁸ However, these markers only partially elucidate the underlying biological and genetic mechanisms of AF and HF, necessitating further refinement and investigation. In this paper, we screened out the core genes of AF group and HF group by Lasso regression analysis and RF model analysis and further identified four biomarkers highly correlated with AF and HF, namely, GLUL, NCF2, S100A12, and SRGN. Immunoinfiltration analysis revealed that GLUL, NCF2, and S100A12 exhibited the strongest correlation with neutrophils in both AF and HF groups. Furthermore, these four genes predicted 101 miRNAs, and GLUL, NCF2, and S100A12 predicted a total of 10 potential therapeutic agents.

With the emergence of multi-omics technology, an increasing number of studies have revealed biomarkers that can be utilized for early diagnosis of AF and HF through integrated analysis of gene research results and transcriptome. This has enhanced researchers' understanding of the pathogenesis of AF and HF. For example, Zhang et al. have shown that CXCR4 and TYROBP are involved in the occurrence of AF, and the mechanism of action may be related to the activation of inflammatory pathways.¹⁹ The PDE3A and GSK3B genes, as identified by Zhou et al., exhibit a significant association with the risk of AF in the Chinese population, providing a new theoretical basis for studying the pathogenesis of AF.²⁰ Based on WGCNA and machine learning techniques, Zhu et al. found that NPPA, OMD, and PRELP could be used as diagnostic biomarkers for DCM combined with HF.²¹ These findings may help us better understand the pathogenesis of AF and HF, as well as facilitate the discovery of new diagnostic biomarkers or therapeutic targets. The co-existence of AF and HF poses challenges in treatment, with a poor prognosis for patients. Identification of shared biomarkers and underlying mechanisms between AF and HF could offer novel insights for enhancing patient prognosis. However, there are few studies revealing the common biomarkers and pathogenesis of AF and HF. Zhuang et al. found that five genes, FRZB, SFRP4, ETNPPL, AQP4, and C1orf105, were highly co-expressed in AF and HF patients through transcriptome data analysis.²² In this study, we found four new biomarkers highly correlated with AF and HF, namely, GLUL, NCF2, S100A12, and SRGN, which is different from the results of Zhuang et al. Proteins encoded by GLUL belong to the glutamine synthase family and play a role in acid–base homeostasis, cell signalling, and cell proliferation. The DEGs associated with HF identified by Boyang et al. through analysis of the Omnibus database included GLUL.²³ Huang et al. reported that SRGN may be a new HF target in iDCM patients,²⁴ and our screening results were consistent with theirs, finding that GLUL and SRGN were differentially expressed in HF, while the gene was also differentially expressed in AF. Currently, no previous studies have identified the expression and function of GLUL and SRGN in AF, which is also an innovative aspect of this study, as it reveals the relationship between GLUL, SRGN, and AF for the first time. NCF2 encodes neutrophil cytoplasmic factor, and studies have found that NCF2 gene and its genetic variation are associated with susceptibility to a variety of diseases. In addition, NCF2 expression was significantly increased in patients with AF compared with the control group.²⁵ S100A12 is a member of the S100 low molecular weight protein family. It was found that the increased expression of S100A12 was correlated with HF.²⁶ It has been reported that S100A12 can be used as the pivotal gene of AF.²⁷ The above research content further supports our results, indicating a strong correlation between the biomarkers GLUL, NCF2, S100A12, and SRGN with AF and HF. These findings indicate that the screened biomarkers have a great significance in diagnosing AF and HF. Gellen et al. found that S100A12 may be a biomarker for HF risk assessment in patients with diabetes.²⁶ This study further indicated that the presence of diabetes in HF patients may affect the expression of S100A12 in HF. In addition, clinical factors, such as age, gender, medication treatment, in AF and HF patients can also impact the expression of GLUL, NCF2, S100A12, and SRGN. This is an area that requires further supplementation and refinement in future studies. In recent years, many scholars have found that immune infiltration plays an indispensable role in the development of diseases including AF and HF. As an important part of the immune system, immune cells play an important role in the immune process. In this study, we observed a significant positive correlation between GLUL, NCF2, and S100A12 with neutrophils in the AF group. Additionally, SRGN exhibited the strongest association with T cells CD4 memory resting. Furthermore, GLUL, NCF2, S100A12, and SRGN exhibited the strongest positive correlation with neutrophils in the HF group. Tang et al. found that KLF2 participated in HF progression by regulating neutrophil activation.²⁸ Friedrichs et al. demonstrated that neutrophils have the capability to induce AF.²⁹ LIAS was also found to be positively correlated with T cells CD4 memory resting in AF. Based on the analysis of previous studies, we speculated that the key genes we screened may participate in the development of AF and HF by regulating the biological functions of neutrophils and resting memory CD4⁺ T cells, but their specific regulatory mechanisms need to be further studied and discussed.

In summary, our findings reveal four potentially key genes: GLUL, NCF2, S100A12, and SRGN. These genes can be used as highly correlated biomarkers for AF and HF, providing new insights into the co-pathogenesis of AF and HF and providing therapeutic targets for patients with co-existing AF and HF. However, our study also has limitations as the four potential key genes were only screened and studied using bioinformatics methods, and there was a lack of gene expression verification in clinical samples. In future studies, we will collect clinical samples from AF and HF patients for relevant verification. In addition, the pathogenesis of AF and HF is caused by the interaction between genetic factors and their environment. Our study lacks analysis and discussion of other contributing factors. Future studies should focus on addressing this gap to further enhance our research findings.

Conflict of interest

The authors declare that there are no conflicts of interest.

Funding

This research was funded by the Special Project for High-quality Development of Shanxi Province's Big Health Industry, DJKZXKT2023116.

References

Baman JR, Passman RS. Atrial fibrillation. JAMA 2021;325:2218. doi:10.1001/jama.2020.23700

Mulder BA, Rienstra M, Van Gelder IC, Blaauw Y. Update on management of atrial fibrillation in heart failure: a focus on ablation. Heart (British Cardiac Society) 2022;108:422‐428. doi:10.1136/heartjnl‐2020‐318081

Tomasoni D, Adamo M, Lombardi CM, Metra M. Highlights in heart failure. ESC Heart Failure 2019;6:1105‐1127. doi:10.1002/ehf2.12555

Gutierrez C, Blanchard DG. Diagnosis and treatment of atrial fibrillation. Am Fam Physician 2016;94:442‐452.

Sagris M, Vardas EP, Theofilis P, Antonopoulos AS, Oikonomou E, Tousoulis D. Atrial fibrillation: pathogenesis, predisposing factors and genetics. Int J Mol Sci 2021;23: [eLocator: 6]. doi:10.3390/ijms23010006

Middeldorp ME, Kamsani SH, Sanders P. Obesity and atrial fibrillation: prevalence, pathogenesis, and prognosis. Prog Cardiovasc Dis 2023;78:34‐42. doi:10.1016/j.pcad.2023.04.010

Lubitz SA, Yin X, Fontes JD, Magnani JW, Rienstra M, Pai M, et al. Association between familial atrial fibrillation and risk of new‐onset atrial fibrillation. JAMA 2010;304:2263‐2269. doi:10.1001/jama.2010.1690

Ragab AAY, Sitorus GDS, Brundel B, de Groot NMS. The genetic puzzle of familial atrial fibrillation. Front Cardiovasc Med 2020;7:14. doi:10.3389/fcvm.2020.00014

Ezeani M, Prabhu S. PI3K(p110α) as a determinant and gene therapy for atrial enlargement in atrial fibrillation. Mol Cell Biochem 2023;478:471‐490. doi:10.1007/s11010‐022‐04526‐w

Liu D, Yang M, Yao Y, He S, Wang Y, Cao Z, et al. Cardiac fibroblasts promote ferroptosis in atrial fibrillation by secreting Exo‐miR‐23a‐3p targeting SLC7A11. Oxid Med Cell Longev 2022;2022. doi:10.1155/2022/3961495

Liu P, Sun H, Zhou X, Wang Q, Gao F, Fu Y, et al. CXCL12/CXCR4 axis as a key mediator in atrial fibrillation via bioinformatics analysis and functional identification. Cell Death Dis 2021;12:813. doi:10.1038/s41419‐021‐04109‐5

King KC, Goldstein S. Congestive heart failure and pulmonary edema [M]. StatPearls. Treasure Island (FL) ineligible companies. Disclosure: Samuel Goldstein declares no relevant financial relationships with ineligible companies. StatPearls Publishing.

Rosik J, Szostak B, Machaj F, Pawlik A. Potential targets of gene therapy in the treatment of heart failure. Expert Opin Ther Targets 2018;22:811‐816. doi:10.1080/14728222.2018.1514012

Chen X, Zhang Q, Zhang Q. Predicting potential biomarkers and immune infiltration characteristics in heart failure. Math Biosci Eng: MBE 2022;19:8671‐8688. doi:10.3934/mbe.2022402

Li J, Zhang Y, Lu T, Liang R, Wu Z, Liu M, et al. Identification of diagnostic genes for both Alzheimer's disease and metabolic syndrome by the machine learning algorithm. Front Immunol 2022;13. doi:10.3389/fimmu.2022.1037318

Li X, Xu C, Li Q, Shen Q, Zeng L. Exploring key genes associated with neutrophil function and neutrophil extracellular traps in heart failure: a comprehensive analysis of single‐cell and bulk sequencing data. Front Cell Dev Biol 2023;11: [eLocator: 1258959]. doi:10.3389/fcell.2023.1258959

Carlisle MA, Fudim M, Devore AD, Piccini JP. Heart failure and atrial fibrillation, like fire and fury. JACC Heart Failure 2019;7:447‐456. doi:10.1016/j.jchf.2019.03.005

Oikonomou E, Zografos T, Papamikroulis GA, Siasos G, Vogiatzi G, Theofilis P, et al. Biomarkers in atrial fibrillation and heart failure. Curr Med Chem 2019;26:873‐887. doi:10.2174/0929867324666170830100424

Zhang YF, Meng LB, Hao ML, Li XY, Zou T. CXCR4 and TYROBP mediate the development of atrial fibrillation via inflammation. J Cell Mol Med 2022;26:3557‐3567. doi:10.1111/jcmm.17405

Zhou Z, Wang Y, Li X, Zhang Y, Yuan L, Chen D, et al. PDE3A and GSK3B as atrial fibrillation susceptibility genes in the Chinese population via bioinformatics and genome‐wide association analysis. Biomedicine 2023;11: [eLocator: 908]. doi:10.3390/biomedicines11030908

Zhu Y, Yang X, Zu Y. Integrated analysis of WGCNA and machine learning identified diagnostic biomarkers in dilated cardiomyopathy with heart failure. Front Cell Dev Biol 2022;10: [eLocator: 1089915]. doi:10.3389/fcell.2022.1089915

Zhuang Y, Qiao Z, Bi X, Han D, Jiang Q, Zhang Y, et al. Screening and bioinformatics analysis of crucial gene of heart failure and atrial fibrillation based on GEO database. Medicina (Kaunas) 2022;58: doi:10.3390/medicina58101319

Boyang C, Yuexing L, Yiping Y, Haiyang Y, Xufei Z, Liancheng G, et al. Construction and analysis of heart failure diagnosis model based on random forest and artificial neural network. Medicine 2022;101: [eLocator: e31097]. doi:10.1097/MD.0000000000031097

Huang H, Luo B, Wang B, Wu Q, Liang Y, He Y. Identification of potential gene interactions in heart failure caused by idiopathic dilated cardiomyopathy. Med Sci Monit: Int Med J Exp Clin Res 2018;24:7697‐7709. doi:10.12659/MSM.912984

Zheng PF, Zhou SY, Zhong CQ, Zheng ZF, Liu ZY, Pan HW, et al. Identification of m6A regulator‐mediated RNA methylation modification patterns and key immune‐related genes involved in atrial fibrillation. Aging 2023;15:1371‐1393. doi:10.18632/aging.204537

Gellen B, Thorin‐Trescases N, Thorin E, Gand E, Ragot S, Montaigne D, et al. Increased serum S100A12 levels are associated with higher risk of acute heart failure in patients with type 2 diabetes. ESC Heart Failure 2022;9:3909‐3919. doi:10.1002/ehf2.14036

Qu Q, Sun JY, Zhang ZY, Su Y, Li SS, Li F, et al. Hub microRNAs and genes in the development of atrial fibrillation identified by weighted gene co‐expression network analysis. BMC Med Genomics 2021;14:271. doi:10.1186/s12920‐021‐01124‐5

Tang X, Wang P, Zhang R, Watanabe I, Chang E, Vinayachandran V, et al. KLF2 regulates neutrophil activation and thrombosis in cardiac hypertrophy and heart failure progression. J Clin Invest 2022;132: doi:10.1172/JCI147191

Friedrichs K, Adam M, Remane L, Mollenhauer M, Rudolph V, Rudolph TK, et al. Induction of atrial fibrillation by neutrophils critically depends on n11b/CD18 integrins. PLoS ONE 2014;9: [eLocator: e89307]. doi:10.1371/journal.pone.0089307