Exploring the molecular mechanisms and shared potential drugs between rheumatoid arthritis and arthrofibrosis based on large language model and synovial microenvironment analysis

Abstract

Rheumatoid arthritis (RA) and arthrofibrosis (AF) are both chronic synovial hyperplasia diseases that result in joint stiffness and contractures. They shared similar symptoms and many common features in pathogenesis. Our study aims to perform a comprehensive analysis between RA and AF and identify novel drugs for clinical use. Based on the text mining approaches, we performed a correlation analysis of 12 common joint diseases including arthrofibrosis, gouty arthritis, infectious arthritis, juvenile idiopathic arthritis, osteoarthritis, post infectious arthropathies, post traumatic osteoarthritis, psoriatic arthritis, reactive arthritis, rheumatoid arthritis, septic arthritis, and transient arthritis. 5 bulk sequencing datasets and 4 single-cell sequencing datasets of RA and AF were integrated and analyzed. A novel drug repositioning method was found for drug screening, and text mining approaches were used to verify the identified drugs. RA and AF performed the highest gene similarity (0.77) and functional ontology similarity (0.84) among all 12 joint diseases. We figured out that they share the same key pathogenic cell including CD34 + sublining fibroblasts (CD34-SLF) and DKK3 + sublining fibroblasts (DKK3-SLF). Potential therapeutic target database (PTTD) was established with the differential expressed genes (DEGs) of these key pathogenic cells. Based on the PTTD, 15 potential drugs for AF and 16 potential drugs for RA were identified. This work provides a new perspective on AF and RA study which enhances our understanding of their pathogenesis. It also shed light on their underlying mechanism and open new avenues for drug repositioning studies.

Introduction

Synovial microenvironment plays an indispensable role in inflammatory joint disease^1–3. It comprises a lining layer rich in lining fibroblasts (LLF) and macrophages, supported by a sublining connective tissue layer containing sublining fibroblasts (SLF) and an intricate vascular network. During joint inflammation, significant alterations in cellular composition occur within this microenvironment⁴, with the activation of macrophages and fibroblasts leading to cytokine production. This inflammatory process associated with the synovial tissue has been already observed in various forms of joint diseases including gouty arthritis, infectious arthritis, juvenile idiopathic arthritis, osteoarthritis, post infectious arthropathies, post traumatic osteoarthritis, psoriatic arthritis, rheumatoid arthritis, etc. ^5–7.

Despite extensive research, most studies have focused on individual diseases⁸, with limited systemic investigation into joint disorders from the perspective of the synovial microenvironment. This limitation has hindered our comprehensive understanding of joint diseases. Moreover, as the literature on joint diseases rapidly expands, it becomes unfeasible to manually read and summarize all of it. Consequently, these demands prompt us to employ a natural language processing (NLP) algorithm to extract information from the abundant literature.

Recently, ChatGPT, a large language model that has opened a new era in NLP and achieved ideal results in various tasks^9,10. This model can precisely extract and capture key information or knowledge from a large scale of unstructured textual data without any pre-training¹¹. Its applications span bioinformatics, computational biology, and systems biology, based on data mining, machine learning, and NLP techniques^12–14. The extracted information often leads to new discoveries and the formulation of novel hypotheses through the construction of knowledge networks and integration with knowledge graphs^15,16.

In this study, we performed a correlation analysis of various joint diseases using the GPT-3.5-Turbo model and revealed the highest similarity between rheumatoid arthritis (RA) and arthrofibrosis (AF) across 12 common joint diseases. Rheumatoid arthritis (RA) is a long-term autoimmune disease which causes severe inflammation of multiple joints. Arthrofibrosis (AF) refers to a fibrotic deposition complication after surgery and trauma that affects many joints. Both diseases exhibit similar symptoms and histopathological features, including chronic synovial hyperplasia, joint stiffness, pain, and excessive collagen and fibrous tissue production within the synovium. Both conditions are marked by synovial linings infiltrated with diverse immune cells and heterogeneous fibroblasts.

In fact, RA has been widely correlated with multiple fibrosis diseases in different organs, including idiopathic pulmonary fibrosis (IPF), cystic fibrosis (CF), liver fibrosis, interstitial lung disease (ILD), myocardial fibrosis, subretinal fibrosis, and lobular fibrosis in labial salivary glands^39–43. Up to 20–40% of RA patients develop pulmonary fibrosis, and 3% develop liver fibrosis. In recent years, several studies have also noted the association between RA and AF. However, no specialized research has been conducted on their correlation.

Therefore, this study aims to perform a comprehensive analysis between RA and AF and identify novel drugs for clinical use. Based on bulk sequencing and single-cell sequencing data, we demonstrated a presently unexplored tight connection and similarity between RA and AF. We explored the common pathogenic cell types and established a potential target database based on differentially expressed genes (DEGs). Additionally, we developed a novel drug repositioning method and identified drugs for RA and AF treatment.

Methods

Diseases similarity analysis

We applied the GPT-3.5-Turbo model for disease-related genes extraction^9,10. In detail, we downloaded 373,644 literatures of 12 different common joint diseases from PubMed, including arthrofibrosis, gouty arthritis, infectious arthritis, juvenile idiopathic arthritis, osteoarthritis, post infectious arthropathies, post traumatic osteoarthritis, psoriatic arthritis, reactive arthritis, rheumatoid arthritis, septic arthritis, and transient arthritis (Supplementary Table 1). Genes information was extracted from the abstracts. Then we performed correlation analysis and network analysis with R package networkD3 (Version 0.4, https://github.com/reymont/networkD3) and GOSemSim (Version 2.16.1) with Resnik’s method¹⁷.

Single-cell data collection and integration

We have enrolled 4 different knee joint synovium single-cell RNA sequencing datasets of rheumatoid arthritis and 1 dataset of arthrofibrosis^18–21. These data were processed and integrated as the previous study described²². In details, all of these datasets were processed with a standard Seurat (Version 3.3) for R (V.3.6) protocol and integrated with Harmony(Version 0.1)^23,24. We obtained a total of 51,430 synovial cells, with 7621 cells from arthrofibrosis patients, 32,296 from rheumatoid arthritis patients, and 11,513 from healthy controls. We used the first 20 PCs for dimensional reduction and cell clustering. Cell annotation was performed manually based on cell markers from the CellMarker 2.0 database^25,26.

Bulk sequencing data collection and preprocessing

All the bulk sequencing data were downloaded from the NCBI Gene Expression Omnibus web resource (GEO, http://www.ncbi.nlm.nih.gov/geo) including 4 rheumatoid arthritis datasets (GSE55457, GSE55584, GSE55235, GSE12021) and one arthrofibrosis dataset (GSE135854)^27–29. These datasets were integrated and normalized to remove the batch effect. The differential expressed genes (DEGs) were calculated with the R package limma (Version 3.46.0, https://CRAN.R-project.org/package = limma) of R software (Version 4.2.0)³⁰. The dimensional reduction analysis was performed with the R package tinyarray (Version 2.2.7, https://CRAN.R-project.org/package=tinyarray).

Deconvolution analysis of bulk RNA sequencing data

The cell type compositions in bulk RNA sequencing data were calculated with deconvolution analysis by R packages MuSiC (Version 0.2.0)³¹. This cell types information was obtained from the integrated and annotated single-cell RNA sequencing data. We used the standard protocol and default parameter setting in this analysis.

Gene functional enrichment analysis

We performed Gene ontology (GO) and KEGG analysis with the R package ClusterProfiler (Version 4.0)³². Gene signatures analysis was conducted by the R packages UCell (Version 1.3)³³ and irGSEA (Version 1.1.3 https://github.com/chuiqin/irGSEA/). Common gene Venn plots were performed by the Venny (Version 2.1, https://bioinfogp.cnb.csic.es/tools/venny/index.html). The Sankey plot of common KEGG pathways was depicted with the R packages networkD3 (Version 0.4, https://github.com/reymont/networkD3).

Pseudotime trajectory analysis

The R package Monocle 3 (Version 1.2.9) was used to analyze the pseudotime trajectory in fibroblast cells. These fibroblast cells were ordered after the overall ‘trajectory’ of gene expression changes has been identified. Default settings were used for all tasks.

Cell communication analysis

A cell communication analysis of single-cell sequencing data was carried out using the R package CellChat (Version 1.4.0)³⁴, involving 1,939 validated molecular interactions.

Potential therapeutic target database development

We have integrated the differential expressed genes of pathogenic cell types including DKK3-SLF, CD55-SLF, F-Mφ, and IFN-Mφ (|LogFc|> 1, adjust p < 0.05). Then the DEGs were overlapped with bulk sequencing DEGs of AF and RA separately. The common genes of AF and RA were eventually enrolled in the arthrofibrosis potential therapeutic target database (AF-PTTD) and rheumatoid arthritis potential therapeutic target database (RA-PTTD).

Drug repositioning and statistical analysis

Rheumatoid arthritis related drugs and their associated genes were collected by the R package DSEATM (Version 1.0.0) with the Mesh Term ID “D001172”³⁵. There are few studies specific on the arthrofibrosis drugs, so we used the fibrosis drugs and related genes instead (Mesh Term ID: “D005355”). Potential arthrofibrosis target genes and rheumatoid arthritis were integrated with scRNA and bulk RNA differential expressed genes. We can set T to represent the potential target genes set with a length of M.

T = [t₁, t₂,…, t^M].

Set D represent the drug genes with a length of N.

D = [d₁, d₂,…, d^N].

Then the matching score can be represented as follows:

$$\text{Matching}\text{score}=\underset{-\infty }{\overset{\infty }{\int }}\frac{\left|\left(\left\{t_i\left|\text{Log}Fc\right.〉x,i\in \left[1,M\right]\right\}\right)\cap \left(\left(d_j\left|\text{Log}Fc\right.〉x,j\in \left[1,N\right]\right\}\right)\right|}{M+N}\text{d}x$$

Drugs with matching score over 0 were defined as potential drug candidates (PDC), and those with a score over 1 as highly potential drugs (HPD).The smooth curve of matching target gene numbers changing over different log fold change cut-off (logFc) were fitted with locally weighted regression method (LOESS). The number of supporting literatures were extracted from PubMed with the same protocol mentioned above. The supporting literature was defined as the literatures that mentioned the drug in the repositioning target diseases. The correlation scatter plot of the matching score and literature score was depicted by the R packages ggstatsplot (Version 0.9.4)³⁶.

Results

Similarity analysis of 12 different joint diseases

The disease similarity analysis study enrolled 373,644 literatures on 12 common joint diseases (Supplementary Table 1). A total of 3042 disease associated genes were extracted and a network was constructed by the top 5 genes with the highest-ranking score (Fig. 1A). Rheumatoid arthritis and arthrofibrosis shared the highest number of common genes across the 12 diseases, with the gene similarity calculation revealing a correlation coefficient of 0.77 (Fig. 1B). GO term similarity calculation was consistent with the gene similarity (0.84) (Fig. 1C). These results indicated that the pathogenesis of rheumatoid arthritis may resembles arthrofibrosis’.

Figure 1.

Similarity analysis of 12 different joint diseases. (A) Top ranking diseases-associated gene network. (B) Heatmap of genes similarity between rheumatoid arthritis and arthrofibrosis. (C) Heatmap of GO terms similarity between rheumatoid arthritis and arthrofibrosis.

Cellular contribution analysis of rheumatoid arthritis and arthrofibrosis

A total of 51,430 synovial cells from 5 different single-cell datasets (4 RA and 1 AF) were successfully integrated together (Fig. 2A). Then these cells were manually annotated into seven different cell types including fibroblast cells, myofibroblast cells, endothelial cells, mononuclear phagocytes, mast cells, T cells, and B cells (Fig. 2B, Supplementary Table 2). The number of fibroblast cells was increased in both rheumatoid arthritis and arthrofibrosis synovial microenvironment (Fig. 2C). Myofibroblasts showed a pronounced increase in AF compared to RA and control groups, while immune cells such as mononuclear phagocytes, T cells, and B cells were more prevalent in RA.

Figure 2.

Cellular contribution analysis of rheumatoid arthritis and arthrofibrosis. (A, B) Unsupervised clustering of single-cell RNA sequencing, visualized by uniform manifold approximation and projection (UMAP) showing major clusters. Each point in the figure represents a single cell. (C) Cell proportion calculation among rheumatoid arthritis patients group, arthrofibrosis group and health control group. (D) Dimension reduction of bulk sequencing data by PCA. (E) Violin plot of estimated cell proportion including myofibroblast, fibroblast, mononuclear phagocytes, and T cells. (F) Semi-Violin plot of 5 common hallmark gene sets in rheumatoid arthritis and arthrofibrosis. (G) Venn plot of KEGG pathway of rheumatoid arthritis bulk sequencing data (RA bulk), rheumatoid arthritis single cell RNA sequencing data (RA sc), arthrofibrosis bulk sequencing data (AF bulk), and arthrofibrosis single cell RNA sequencing data (AF bulk). (H) Sankey plot of common KEGG pathway and common genes in rheumatoid arthritis single cell RNA sequencing data (RA sc), arthrofibrosis bulk sequencing data (AF bulk), and arthrofibrosis single cell RNA sequencing data (AF bulk).

Bulk sequencing analysis also revealed the similarity of RA and AF. Principal component analysis (PCA) of bulk tissue data showed RA and AF clustering closely compared to controls (Fig. 2D). The deconvolution calculation of cell type proportion confirmed the increased fibroblast cell count in both RA and AF (Fig. 2E). Changes in myofibroblast cells, mononuclear phagocytes, and T cells were consistent with single-cell data. B cells were not detected in this analysis due to the small sample size.

We also revealed that RA and AF share the key biological processes (Fig. 2F), with elevated gene sets related to inflammatory response, apical junction, and TNF-α signaling pathways. KEGG analysis figured out that they share as many as 25 common pathways (Fig. 2G). This was a relatively large proportion. The Sankey plot depicted the most important common pathways including the MAPK signaling pathway, Cytokine-cytokine receptor interaction, Toll-like receptor signaling pathway, TGF-β signaling pathway, ECM receptor interaction, GAP junction, and regulation of actin cytoskeleton (Fig. 2H). These are all important in the pathogenesis of both AF and RA.

Fibroblast subcluster analysis

Based on the previous studies^22,37, fibroblasts can be further annotated into 3 subclusters including CD55 + lining fibroblasts (CD55-LLF), CD34 + sublining fibroblasts (CD34-SLF), and DKK3 positive sublining fibroblasts (DKK3-SLF) (Fig. 3A). CD55-LLF fibroblasts located in the synovium lining layer (LL). It expressed high level of CD55, PRG4, MMP3, FN1, and low level of THY1 (Fig. 3B). CD34-SLF, found in the sub-lining layer (SL), showed high expression of CD34, CXCL12, CXCL14, CCL2, and the highest level of THY1. The third fibroblast cell type, DKK3-SLF was also in sub-lining layer which expressed high levels of DKK3, POSTN, and intermediate levels of THY1. The DKK3-SLF and CD34-SLF are pathogenic cells that both increased in RA and AF (Fig. 3C).

Figure 3.

Fibroblast subcluster analysis. (A) Dimensional reduction and functional enrichment of fibroblast subcluster cells. (B) Marker genes of different fibroblast subcluster (C) Bar plot of different cell proportion in single cell data. (D) Violin plot of estimated cell proportion of bulk sequencing data. (E) Different cell stage proportion of fibroblast subcluster cells. (F) Pseudotime trajectory analysis of fibroblast cells. The black arrow indicates the differentiation direction of fibroblasts. (G) The signaling pattern of AF and RA.

The cell proportion estimation of bulk sequencing data supported these findings (Fig. 3D), showing higher proportions of DKK3-SLF and CD34-SLF in RA and AF compared to controls. Moreover, active proliferation of DKK3-SLF in both RA and AF was indicated by the increased proportions in S and G2M stages (Fig. 3E).

The pseudotime analysis showed a clear trajectory of CD55-LLF developed into CD34-SLF and DKK3-SLF (Fig. 3F). CD34-SLF was majorly increased in RA and DKK3-SLF in AF. Cell chat analysis also showed that DKK3-SLF and CD34-SLF are the major contributors in cell interaction (Fig. 3G). These results revealed that the pathogenic fibroblast cells were very similar in AF and RA, both in composition and behavior.

Mononuclear phagocytes subcluster analysis

The integrated mononuclear phagocytes were also divided into 7 different subclusters including conventional type 1 dendritic cells (cDC1), conventional type 2 dendritic cells (cDC2), M1-like fibrotic macrophage (M1-like F-Mφ), M2-like fibrotic macrophage (M2-like F-Mφ), interferon-stimulated macrophages (IFN-Mφ), S100A8/9 high macrophages (S100A8/9-Mφ), inflammatory macrophages (I-Mφ), immune regulated macrophages (IR-Mφ), transitional macrophages (T-Mφ) and others (Fig. 4A). The key pathogenic cell types F-Mφ and IFN-Mφ were elevated in both RA and AF (Fig. 4B). Functional annotation showed that F-Mφ were highly associated with extracellular matrix and collagen trimer (Fig. 4C), while IFN-Mφ were correlated with leukocyte migration and neutrophil activation.

Figure 4.

Mononuclear phagocytes subcluster analysis. (A) Dimensional reduction of mononuclear phagocytes subcluster cells. (B) Bar plot of different cell proportion in single cell data. (C) Functional enrichment of different mononuclear phagocytes subcluster cells. (D) M1/M2 polarization score of different mononuclear phagocytes subcluster cells. (E) M1 polarized cell proportion of different groups. (F) String plot of mononuclear phagocytes interaction of AF and RA. (G) Dimensional reduction of mononuclear phagocytes interaction in AF and RA.

To address the commonly used M1/M2 classification of macrophage, we calculated the M1/M2 polarization score using previously defined M1- and M2-associated marker genes³⁸. F-Mφ and IFN-Mφ were enriched in M1 genes while other macrophage subsets such as IR-Mφ and cDC1 expressed mostly M2 genes (Fig. 4D). AF and RA exhibited similar M1 polarized cell proportions, which far exceed the control group (Fig. 4E).

We also performed cell chat analysis in the mononuclear phagocytes subcluster. The cell interaction patterns of AF and RA were also very similar to each other (Fig. 4F and G). Moreover, the interaction analysis between fibroblasts and macrophages indicated extensive crosstalk involving TGF-β, PDGF and collagen related pathways (Supplementary Fig. 1A, B). Compared with the control group, RA and AF are similar in this cell crosstalk.

T cells subcluster analysis

T cells were further annotated into 9 different subclusters including T follicular helper (Tfh), natural killer T cells (NKT), Naive CD8 + T cells, Naive CD4 + T cells, CD8 + effector T cells, CD4 + T helper cells 1 (Th1), CD4 + T helper cells 2 (Th2), CD4 + T helper cells 17(Th17), and CD4 + regulatory T cells (Treg) (Fig. 5A). T cells subcluster similarity analysis showed that the T cells in AF resembled those in RA (Fig. 5B). Th1 and Th2 were both increased in RA and AF (Fig. 5C). Cell chat analysis also revealed the similarity of cell interactions between AF and RA (Fig. 5D and E).

Figure 5.

T cells subcluster analysis. (A) Dimensional reduction of T cell subcluster cells. (B) Heatmap of T cells subcluster similarity analysis. (C) Bar plot of different T cells proportion in single cell data. (D) String plot of T cells interaction of AF and RA. (E) Dimensional reduction of T cells interaction in AF and RA.

Drug repositioning of AF and RA

In this study, we have developed a novel drug repositioning method based on the integration of common differential expressed genes of pathogenic cell types including DKK3-SLF, CD55-SLF, F-Mφ and IFN-Mφ. There were 395 rheumatoid arthritis related drugs and 266 fibrosis related drugs enrolled in the screening (Supplementary Table 3 and Supplementary Table 4). 15 of the rheumatoid arthritis related drugs were successfully repositioned in AF based on the matching score calculation (Supplementary Table 5). The matching score and target gene numbers of each potential drug candidate were shown in Fig. 6A. The top 6 drugs selected as the highly potential drugs (HPD) for AF treatment included Triazines, Isoxazoles, Anti-Inflammatory Agents (Non-Steroidal), Propionates, Auranofin, and Penicillamine (Fig. 6B). The matching target genes of top 6 drugs with different changes (logFc) were depicted in Fig. 6C.

Figure 6.

RA drugs repositioned in AF. (A) Scatter plot of matching score and target gene numbers of each drug. Red point represents the match score over 1 while the blue points were less than 1. (B) Fitting plot of highly potential drugs’ matching gene numbers changing over logFc. (C) Dot plot of match gene of 6 top screening drugs. (D) Scatter plot of supporting literature and all drug-related literatures. (E) Correlation scatter plot of literature score and match score.

To validate these drugs, we searched PubMed to get the number of literatures that support the repositioning of each drug. The literature score was calculated with the number of all literature and supporting literature shown in Fig. 6D. The correlation analysis revealed that the matching score was highly correlated with the literature score (R = 0.86, Fig. 6E). This confirmed that the matching scores were reliable criteria for drug repositioning evaluation.

Meanwhile, 16 fibrosis-related drugs were successfully repositioned in RA (Supplementary Table 6). The matching score and target gene numbers of each drug were also shown in Fig. 7A. The top 6 HPD were Emodin, Histone Deacetylase Inhibitors, Adenosine A2 Receptor Agonists, Benzamide, Azo Compounds, 2-methyl-2H-pyrazole-3-carboxylic acid (2-methyl-4-o-tolylazophenyl) amide (Fig. 7B and C). Literature studies confirmed the high correlation of matching score and literature score (Fig. 7D and E).

Figure 7.

Fibrosis drugs repositioned in RA. (A) Scatter plot of matching score and target gene numbers of each drug. Red point represents the match score over 1 while the blue points were less than 1. (B) Fitting plot of matching target gene numbers changing over logFc. (C) Dot plot of match gene of 6 top screening drugs. (D) Scatter plot of supporting literature and all drug-related literatures. (E) Correlation scatter plot of literature score and match score.

Discussion

This is the first study to describe a comprehensive connection between rheumatoid arthritis and arthrofibrosis, providing important insights into their pathogenesis and guides for clinical treatment.

Fibroblasts were dominant cells in both RA, AF and normal synovium^44,45. Our study showed that the DKK3-SLF and CD34-SLF were the most important cell type in pathogenesis of both RA and AF, with increased proliferation in both conditions. Mizoguchi and colleagues reported that the DKK3-SLF was associated with severe synovitis and synovial hypertrophy by ultrasound¹⁹. Rapheal et al. also found a positive correlation between DKK3-SLF and clinical parameters of RA including visual analog scale (VAS), C-reactive protein level (CRP) and tender joint count (TJC)²². These may reveal that the DKK3-SLF was the key effector cells in the synovial hypertrophy. Based on the expression of chemotactic factors and inflammatory cytokines, CD34-SLF played an important role in monocyte recruitment and regulation of inflammation²².

Moreover, RA and AF show similarities in mononuclear phagocytes subcluster analysis. It was widely accepted that the M1 polarized macrophages secrete principally proinflammatory cytokines such as TNF-α and IL-1, which contribute to the disease progression both in AF and RA^27,46,47. Our study confirmed that AF and RA have the relatively same M1 polarized proportion, which is far beyond the healthy control group.

Recently, many studies are focusing on the crosstalk between the synovial fibroblasts and macrophages, which might be a key to understand the mechanism of these two diseases^48–50. The pathogenic fibroblast cell types such as DKK3-SLF and CD34-SLF produce excessive cytokines and chemokines such as IL-6 and CCL2. This may contribute to the M1 polarization and increased TGF-β and PDGF secretion. These products amplified the inflammatory response of both macrophages and fibroblasts, which led to fibrotic pathogenesis and synovium hyperplasia. Such pathological circuits have been widely accepted in many fibrosis diseases such as lung fibrosis and may also be involved in synovial fibrosis of RA and AF as well^51,52.

The novel drug repositioning method employed in this study, based on DEGs of key pathogenic cell types, offers a more precise approach to drug screening. The identified drugs have shown potential in clinical settings, with some already used for RA and AF treatment. For example, emodin was widely reported to ameliorate rheumatoid arthritis^53–58. Histone Deacetylase Inhibitors was believed as the new hope for rheumatoid arthritis^59–64. Silymarin was also reported to reduce the clinical syndromes in RA patients based on a non-randomized single-arm clinical trial⁶⁵. Auranofin and penicillamine were both used for lung fibrosis and liver fibrosis prevention and treatment^66–72. Azathioprine and Sulfhydryl Compounds can ameliorate the idiopathic pulmonary fibrosis^73–76. Isoxazoles elicited an increase in adult mouse myocardial cell cycle activity while concurrently mitigating myocardial fibrosis ⁷⁷. The positive correlation between matching scores and supporting literature further validates the reliability of this repositioning method.

However, several limitations should be acknowledged. First, the study primarily focuses on the transcriptomic data of the synovial microenvironment in RA and AF. These findings may not be fully generalizable to all RA and AF patients due to variations in disease presentation, patient demographics, genetic backgrounds, etc. Second, extracting data from existing literature may introduce bias if the extracted information is incomplete or skewed towards certain research trends. Lastly, although the study identifies potential therapeutic targets and drugs through computational analysis, it lacks experimental validation of these findings. Without laboratory or clinical confirmation, the practical applicability of these results remains uncertain. Future research could involve expanding the datasets, incorporating more diverse patient samples, and conducting experimental validations to confirm the computational predictions.

Conclusion

This study uncovers a previously unexplored connection between RA and AF through integrated sequencing data and text mining. We also developed a potential targets database and identified several potential drugs based on the knowledge. This work provides a new perspective on AF and RA study which enhances our understanding of the pathogenesis.

Data and code availability

These data and code can be found at: https://github.com/chenxi199506/RAandAF.

Author contributions

S.T.C. and X.C designed the experiments; X.C. and Z.Q.W. performed the experiments; X.C., C.L., and Z.Q.W. analyzed the data; X.C. wrote the manuscript; S.T.C. and Y.S.S. revised the manuscript. All authors read and approved the final manuscript.

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-69080-5.