Large-scale transcriptomic data mining using explainable XGBoost and SHAP reveals shared biomarkers and molecular mechanisms between type-2 diabetes and triple-negative breast cancer for drug repurposing Authors: Md Feroj Ahmed, Ovi Sharma, Tasfia Noor, Md Abdul Latif, Md Foysal Ahmed, Moumita Bashak, Dewan Mumdood Ahmed, Md Sanoar Hossain, Md Nurul Haque Mollah Published: 2026 ABSTRACT AbstractType 2 diabetes (T2D) is considered as a risk factor of triple-negative breast cancer (TNBC). So, there is a significant chance of their co-existence. The management of TNBC with T2D becomes more complex than without T2D due to the conflict of therapies, since some T2D drugs may have bad impact on TNBC and vice-versa. Beside this, drug-drug interaction due to the polypharmacy of multiple drugs, during a one-drug one-disease strategy, may create toxicity or side effect to the patients who are suffering from both diseases simultaneously. Therefore, it is required to explore effective unique drugs as the same treatment for both diseases. This study attempted to contribute in this issue. At first, we identified 36 shared differentially expressed genes (sDEGs) that can separate both TNBC and T2D patients from the control group through integrated transcriptomics analysis. Then top-ranked four sDEGs (S100A9, CIRBP, USP10, and PSMD1) were detected as the overlapping dysregulated shared key-genes (sKGs) through the protein-protein interaction (PPI) network analysis and filtering with a machine learning (ML) approach. The gene regulatory network analysis revealed three key transcriptional (TFs proteins) and post-transcriptional (micro-RNAs) factors of sKGs. The enrichment analysis of sKGs with the GO-terms and KEGG-pathways revealed some crucial molecular functions, biological processes, cellular components, and pathways as the key pathogenetic mechanisms for the development and progression of both diseases. Finally, we recommended sKGs-guided two repurposable common candidate drugs (tepotinib and ursodiol) for both diseases by molecular docking, ADME/T analysis and MD simulation studies. Thus, the results of this study could provide useful insights for researchers and medical professionals for improving the diagnostic and therapeutic strategies for the treatment TNBC with T2D as comorbidity.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s13040-026-00575-9.Keywords: Type 2 diabetes, Triple-negative breast cancer, Machine learning–based SHAP analysis, Shared key genes and pathogenetic processes, Drug repurposing, Statistics and bioinformatics analysis Introduction Cancer remains an increasing global health concern, and breast cancer (BC) is the most commonly diagnosed malignancy among women, accounting for approximately 23% of all cancer cases [1]. Over the past three decades, the global incidence of BC has continued to rise, and it remains the leading cause of cancer-related death among women, despite significant progress in diagnostic methods and therapeutic approaches [1, 2]. Approximately 1.4 million new BC cases are reported worldwide each year [3]. Triple-negative breast cancers (TNBCs) are characterized by a lack of estrogen receptor (ER), progesterone receptor (PR), and HER2 expression [4]. Triple-negative breast cancer (TNBC) accounts for approximately 15–20% of newly diagnosed breast cancers and remains the only subtype without effective targeted therapeutic options [5]. TNBCs typically show a more aggressive disease pattern than other BC subtypes [6]. Women with TNBC have a high risk of recurrence within three years of diagnosis and experience increased mortality for up to five years thereafter. Across all stages, TNBC is associated with poorer survival compared with other BC subtypes, with the worst outcomes observed among non-Hispanic Black women with late-stage disease, who have a five-year relative survival rate of 14% [3]. Conversely, Type 2 Diabetes (T2D) is one of the most common metabolic disorders worldwide [7]. Current epidemiological evidence indicates a concerning rise in the global burden of T2D. According to the International Diabetes Federation, approximately 463 million adults aged 20–79 years were living with diabetes worldwide, a number projected to increase to nearly 700 million by 2045. In 2019 alone, diabetes was responsible for an estimated 4.2 million deaths globally [7]. An individual with diabetes may have compromised immunological function. The immune system’s vulnerability to invasive pathogens in diabetic people is believed to be caused by deficiencies in both the innate and adaptive immune responses, including malfunctioning T cells and neutrophils and macrophages. This vulnerability may lead a diabetes patient to be at high risk of cancer, infection and other diseases [8, 9]. T2D is considered as a risk factor of TNBC [10]. So, there is a significant chance of their co-existence. Multiple epidemiological studies have shown that individuals with T2D have a 20–30% increased risk of developing TNBC and accounts for approximately 50% of BC related mortality [9, 11, 12]. Patients with concurrent diabetes and TNBC exhibit poorer clinical outcomes and reduced quality of life compared with those with either condition alone, primarily due to an increased risk of distant disease relapse that complicates long-term disease management [2, 11]. Data from cross-sectional studies and tumor tissue microarray analyses demonstrate that BC patients with T2D have a higher likelihood of developing hormone receptor–negative tumors, including TNBC, compared with patients without diabetes [11, 13, 14]. In postmenopausal women, BC occurring in the presence of T2D is more often diagnosed at later stages and is frequently associated with aggressive subtypes such as TNBC. Compared with women without diabetes, these patients also face a higher risk of invasive disease and increased BC-related mortality [11, 15]. The management of TNBC with T2D becomes more complex than without T2D due to the conflict of therapies, since some T2D drugs may have bad impact on TNBC and vice-versa [16]. Beside this, drug-drug interaction due to the polypharmacy of multiple drugs, during a one-drug one-disease strategy, may create toxicity or side effect to the patients who are suffering from both diseases simultaneously [10]. Therefore, it is required to explore effective unique drugs as the same treatment for both diseases. However, so far, no significant gene level analysis has been done in this complex condition as well as no researcher yet rigorously explored such type of common/unique drugs for both diseases, though it is becoming more dangerous day by day. So, in this study, we aimed to investigate the underlying pathophysiological association between T2D and TNBC. Furthermore, we found pathways, transcriptional factors, micro-RNAs and suggested effective common/unique drugs as the same treatment for both diseases. Materials and methods Curated transcriptomics profile collections and preprocessing Transcriptomics profile collections To identify the shared key genes associated with TNBC and T2D, publicly available microarray-based gene expression datasets for both diseases were retrieved from the Gene Expression Omnibus (GEO) database of the National Center for Biotechnology Information (NCBI). For TNBC, the datasets GSE27447 [17], GSE38959 [17], GSE37751 [18], GSE43358 [19], and GSE65194 [20] were selected, while GSE25724 [21], GSE9006 [21], GSE29221 [22], GSE29231 [22], and GSE64998 [23] were obtained for T2D. The selected datasets encompass samples originating from distinct populations, tissue sources, and experimental microarray platforms, providing broad biological and technical diversity. The detailed information of the datasets given in the Table 1. Table 1Detailed information about the transcriptomic datasets which have been used in this studyAccession IDs for datasetsDisease nameCountryPlatformCaseControlGSE27447TNBCUSAGPL6244 [HuGene-1_0-st] Affymetrix Human Gene 1.0 ST Array [transcript (gene) version]514GSE38959TNBCJapanGPL4133 Agilent-014850 Whole Human Genome Microarray 4 × 44K G4112F (Feature Number version)3017GSE37751TNBCUSAGPL6244 [HuGene-1_0-st] Affymetrix Human Gene 1.0 ST Array [transcript (gene) version]1447GSE43358TNBCBelgiumGPL570[HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array170GSE65194TNBCFranceGPL570 [HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array4111GSE25724T2DItalyGPL96[HG-U133A] Affymetrix Human Genome U133A Array67GSE9006T2DUSAGPL96 [HG-U133A] Affymetrix Human Genome U133A Array, and GPL97 [HG-U133B] Affymetrix Human Genome U133B Array2448GSE29221T2DIndiaGPL6947 Illumina HumanHT-12 V3.0 expression beadchip1212GSE64998T2DSwedenGPL11532[HuGene-1_1-st] Affymetrix Human Gene 1.1 ST Array [transcript (gene) version]76GSE29231T2DSwedenGPL6947 Illumina HumanHT-12 V3.0 expression beadchip1212 Detailed information about the transcriptomic datasets which have been used in this study Preprocessing In this study, we merged/combined 5 independent datasets for each of TNBC and T2D to increase the sample sizes in the case (T2D/TNBC) and control groups for identifying differentially expressed genes (DEGs) between cases (T2D/TNBC) and control groups, since statistical power increases with the increasing of sample size [24, 25]. The merged TNBC dataset consisted of 107 TNBC and 89 control samples, whereas the merged T2D dataset consisted of 61 T2D and 85 control samples. To remove the batch effect from the merged datasets, we applied several statistical transformations (quantile normalization, log2 transformation) if needed in the micro-array data, and batch effects were corrected using the PyCombat implementation of the ComBat algorithm in Python package with its default parameters [26]. This method applies an empirical Bayes approach to adjust for batch-associated variation while preserving true biological differences [26]. The effectiveness of batch effect correction was evaluated using boxplot and principal component analysis (PCA) analyses, as presented in Fig. S1 and S2. Identification shared differentially express genes (sDEGs) To identify shared differentially expressed genes (sDEGs), we first determined the differentially expressed genes (DEGs) for each disease by comparing disease versus normal samples using the Linear Models for Microarray Data (LIMMA) framework implemented in the R limma package, a widely accepted statistical approach (see Method S1) for detecting DEGs in transcriptomic analyses [27]. For both diseases, we applied the same cutoff criteria to identify DEGs. Upregulated genes were defined as those with an adjusted p-value ≤ 0.05 and a log₂ fold change (log₂FC) ≥2, while downregulated genes were defined by an adjusted p-value ≤ 0.05 and a log₂FC ≤ −2. After identifying DEGs for each disease, we derived shared expression signatures by intersecting the upregulated gene sets from T2D and TNBC to obtain the shared upregulated DEGs. Similarly obtain the shared downregulated DEGs obtained by intersecting the downregulated gene sets of both diseases. These overlapping genes represent statistically robust and biologically consistent expression changes common to both conditions. The combined set of sDEGs may reflect core molecular pathways jointly disrupted in T2D and TNBC. Shared key genes (sKGs) Identification and Validation framework Network-based Identification of topologically significant shared genes (TSSGs) using protein-protein interaction (PPI) analysis A protein-protein interaction (PPI) network was constructed for the sDEGs using the STRING database [28], applying default settings along with the confidence interaction score 0.5 which is slightly greater than medium confidence interaction score for PPI network construction in String database. Then the resulting interaction data were imported into Cytoscape [29] to produce a clearer and more detailed visualization of the network. We ranked all nodes in the PPI network based on shortest paths using six CytoHubba topological metrics, including Degree, Betweenness, EPC, Closeness, MCC and Stress [30]. These network-based rankings were then used to identify the TSSGs. The TSSGs were selected in such a way that they were mostly connected in the six topological measures mentioned above. Machine-learning (ML) selection of shared feature genes (SFGs) via SHAP value ranking To identify Important SFGs from using ML method, we first extracted the expression values of all sDEGs from the integrated T2D and TNBC datasets. All available samples were used during model training to capture the full range non-linear relationships within gene expression patterns. We then applied an extreme gradient boosting (XGBoost) ML model [31] separately for each disease dataset, to distinguish disease samples from controls. Prior to model fitting the final ML models, 5-fold cross-validation was performed to optimize hyperparameters and minimize the risk of overfitting, ensuring reliable and generalizable results for each model. This allowed the model to pick up the genes that carried the strongest and most consistent signal related to the disease state and progression [32, 33]. Following this, we used SHapley Additive exPlanations (SHAP) values to evaluate gene importance and selected the top 15 contributing genes from each disease-specific model [34]. The genes common to both lists were considered the SFGs, reflecting shared molecular features that may play a role in both T2D and TNBC. To further validate the robustness and generalizability of the models, their performance was assessed using independent test datasets. GSE113865 was used to validated ML model for TNBC and GSE38642 for T2D ML model (see Method S2). Determination of shared key genes (sKGs) through gene-set intersection To refine the list of sKGs, we combined results from the PPI-based ranking and the machine-learning gene importance analysis. Only genes consistently identified by both methods were retained as robust sKGs with potential roles in T2D and TNBC. Revealing the shared regulatory network and disease mechanisms of sKGs Transcription factors and miRNAs play central roles in shaping gene expression programs. TFs regulate gene transcription directly at the DNA level, while miRNAs fine-tune gene expression after transcription by controlling mRNA stability and translation [35, 36]. Gene Ontology (GO) analysis provides a structured framework for interpreting gene functions across biological processes (BPs), molecular functions (MFs), and cellular components (CCs), allowing us to uncover coordinated biological themes involved in disease [37]. Pathway enrichment analysis highlights the major signaling and metabolic pathways in which the sKGs are involved, offering deeper mechanistic insight into the shared disease [38]. Identifying which TFs and miRNAs regulate the sKGs helps reveal upstream molecular drivers that may coordinate the shared pathological processes between T2D and TNBC such as metabolic reprogramming, inflammation, or proliferative signaling. Identification of key regulator of sKGs We performed a regulatory network analysis using both transcriptional regulators (TFs) and post-transcriptional regulators (miRNAs) to identify key molecules that control sKG expression at multiple regulatory layers. This allowed us to examine how sKGs are regulated both before and after transcription, providing deeper insight into their upstream regulatory mechanisms. To identify the key TFs that regulate the sKGs, we have constructed TFs-sKGs interaction network from JASPER database which provided experimentally supported binding profiles [39]. Similary, to disclose key post regulators of sKGs, we have constructed miRNAs-sKGs interaction network using TARbase [40], which includes experimentally validated associations reported in the literature. NetworkAnalyst webtool [41] was used to build these interaction networks. Lastly, we have selected the most important TFs and miRNAs based on Degree and betweenness topological measures. Revealing shared disease mechanisms mediated by sKGs To understand the shared biological functions and pathways associated with the sKGs in T2D and TNBC, we carried out GO enrichment and KEGG pathway analyses. These analyses were performed using the, Enrichr [42], and GeneCodis4 [43] web tools. Enrichment significance was assessed using Fisher’s exact test. To correct for multiple testing, the Benjamini-Hochberg false discovery rate (FDR) method was applied. Go terms and pathways with an adjusted p-value < 0.05 were considered statistically significant for both webtool. And significant GO terms and pathways overlapping in both webtool were considered in this study. sKGs-guided drug repurposing for therapeutic Discovery To identify drug molecules which may be able to treat the condition of TNBC associated with T2D, we have gone through drug repurposing procedures. Drug repurposing allows us to uncover new therapeutic uses for existing drugs by systematically evaluating how approved or investigational compounds may target the molecular signatures identified in our study. Data acquisition, structure curation, and preprocessing pipeline To identify potential therapeutic candidates for TNBC associated with T2D, disease-related drugs were collected from DrugBank [44], while sKGs- and TF-associated drugs were obtained from DGIdb [45] (see Table S2 and S3). Previously reported drugs in published paper to treat these conditions separately were also collected (see Table S1). The three-dimensional (3D) structures of selected drug compounds were downloaded from PubChem [46]. The sKGs and key TFs were then assessed using the DoGSiteScorer [47] web tool to evaluate their druggability. Proteins with a druggability score greater than 0.5 were considered to have potential for drug interaction and molecular docking. Subsequently, experimentally resolved 3D structures of these proteins were retrieved from the Protein Data Bank (PDB) [48], and 3D structures predicted models were retrieved from SWISS-MODEL [49] databases. Prior molecular docking, all structures were preprocessed. Drug molecules were optimized through energy minimization, torsion tree generation, and charge assignment using Avogadro [50] and AutoDockTools4 [51]. Protein structures were refined by removing water molecules, heteroatoms, and unwanted chains in Discovery Studio [52], followed by hydrogen addition, Kollman charge assignment, and grid box preparation in AutoDockTools4 [51]. Molecular docking Blind Docking was carried out to estimate the binding affinity of each compound and to predict the most favorable orientation of the ligand within the protein’s active or regulatory site to avoid potential bias in pocket selection. Molecular docking of each preprocessed drug molecule with the target protein structures was performed using AutoDock Vina [53]. After performing molecular docking between all drugs and target proteins, we calculated the average binding affinity scores (BAS) across both rows (proteins) and columns (drugs). Based on these average scores, all drugs were ranked. The top five drugs with the best overall binding affinities were selected for further analysis. Pharmacokinetics and toxicity analysis Many drug candidates fail during development because of poor pharmacokinetics (Absorption, Distribution, Metabolism, and Excretion) or unexpected toxicity, even when they show strong binding affinity in silico or in vitro [54]. To assess the pharmacokinetic properties and potential toxicity of the top-ranked promising drug candidates, we performed ADME/T (Absorption, Distribution, Metabolism, Excretion, and Toxicity) analyses using tools such as Deep-PK [55] and pkCSM [56]. These predictions helped us evaluate how the selected compounds might behave and how safe they may be within a biological system before considering further investigation (see Method S3). Drug likeliness property analysis To further assess the therapeutic potential of the candidate drug molecules, we evaluated their drug-likeness properties using the Lipinski Rule of Five, which is commonly used to predict oral bioavailability and overall suitability as lead compounds. This evaluation was carried out using online tools such as ADMETlab 3.0 [57] and pkCSM [56], which provided predictions of key physicochemical properties including molecular weight, hydrogen-bonding features, and LogP. These tools also assessed medicinal chemistry aspects such as PAINS alerts, Brenk structural flags, and synthetic accessibility. All drug candidates were submitted in SMILES format, retrieved directly from the PubChem database [46]. Validation of the potential drug molecules To check whether our shortlisted drug molecules are truly effective and reliable, we cross-validated them against the key genes (KGs) of each disease. To identify these KGs, we first built separate protein–protein interaction (PPI) networks for TNBC and T2D and analyzed them in Cytoscape [29]. Using the CytoHubba plugin [30], we applied several topological measures (like Degree, EPC, Closeness, and Betweenness) to find the most influential nodes in each network. The top 10 genes that consistently ranked highest across these measures were selected as the key genes for each disease. The 3D structures of these key proteins for both diseases were then collected from databases such as PDB [48], AlphaFold [58], and SWISS-MODEL [49]. Each structure was prepared using the same preprocessing steps described earlier to ensure they were suitable for downstream docking and analysis. Molecular docking between the KGs protein structures and our potential drug candidates was performed using AutoDock Vina [53]. Finally, we analyzed the BAS to assess the strength and stability of each drug–protein interaction, providing insight into how effectively the candidate drugs may influence KGs function at the molecular level. Molecular dynamics analysis Molecular dynamics (MD) simulation is an essential step for validating docking poses because it captures protein-ligand flexibility and mimics realistic atomic motions that cannot be observed in static docking models [59, 60]. MD simulations were carried out for the top-ranking protein-ligand complexes using YASARA [61] with the AMBER14 force field to investigate their stability and behavior over 100 ns time [62]. Prior this, MD simulation we have examined the receptor-ligand complexes, their interactions at residue level in 2D and 3D view to understand the binding poses and other attributes. The complexes showing the most favorable docking results were selected for 100-ns simulations under near-physiological conditions. System trajectories were recorded at 100-ps intervals and analyzed using YASARA’s built-in macros along with SciDAVis(http://scidavis.sourceforge.net/). Binding free energies (ΔG_bind) were estimated throughout the simulation using the MM-PBSA approach implemented in YASARA, allowing us to assess the strength and consistency of interactions within each complex [63, 64]. Workflow of the study The entire workflow of this study is shown below (see Fig. 1). Graphical abstract of the entire workflow Results Identification shared differentially express genes (sDEGs) Differentially expressed genes (DEGs) that showed statistically robust, reproducible, and biologically consistent expression patterns in both T2D and TNBC were identified using the LIMMA statistical framework. In TNBC, the analysis yielded 2,117 upregulated and 914 downregulated genes based on an adjusted p-value < 0.05 and |log2FC|>2. Using the same criteria, the T2D dataset revealed 217 upregulated and 259 downregulated genes. To determine shared transcriptional alterations, we intersected the upregulated gene sets from both diseases, identifying 24 genes consistently upregulated in T2D and TNBC. Applying an analogous comparison to the downregulated gene sets resulted in 12 shared downregulated genes. In total, 36 sDEGs showed in Table S4 were identified which may reflect core molecular pathways jointly disrupted in T2D and TNBC. The Whole procedure illustrated in Fig. 2. Procedure for identification of sDegs from outcome of each disease Shared key genes (sKGs) Identification and Validation framework Network-based Identification of topologically significant shared genes (TSSGs) using protein-protein interaction (PPI) analysis A PPI network was constructed in STRING using the shared DEGs, and the resulting network was visualized and analyzed in Cytoscape. To identify highly influential genes within the network, we applied six widely used topological algorithms implemented in the CytoHubba plugin (Table 2). Genes that consistently ranked among the top performers across these metrics were considered to have strong centrality and functional importance. The 15 genes that appeared most frequently across the six topological rankings were ultimately designated as the TSSGs. Table 2Top ranked genes in six topological measures and top 15 most occurred TSSGsEPCClosenessStressDegreeMCCBetweennessTSSGsOccurrencePCBP2PCBP2PSMD1PCBP2CIRBPPSMD1PCBP26HSPH1MYD88MYD88S100A9USP10MYD88USP106USP10PSMD1PCBP2MYD88HSPH1S100A9S100A86CIRBPHSPH1S100A9USP10SRPK1PCBP2PSMD16PSMD1S100A9ASCC3PSMD1ASCC3ASCC3SRPK16MYD88CIRBPHSPH1HSPH1PCBP2HCCSASCC36ASCC3ASCC3HCCSCIRBPS100A9HSPH1S100A96SRPK1USP10CIRBPASCC3MYD88ARHGEF11HSPH16S100A9SRPK1USP10S100A8S100A8ACOT9MYD886HABP4HABP4ACOT9SRPK1PSMD1USP10CIRBP6CNOT9S100A8SRPK1HCCSHABP4SRPK1HCCS5S100A8HCCSARHGEF11HABP4ALDH1B1CIRBPHABP44HCCSGPX4S100A8ALDH1B1PLAURGPX4GPX44GPX4CNOT9ME2PLAURDHRS12S100A8ME23ME2ALDH1B1GPX4DHRS12CNOT9ME2CNOT93 Top ranked genes in six topological measures and top 15 most occurred TSSGs Machine-learning (ML) selection of shared feature genes (SFGs) via SHAP value ranking We have extracted the expression values of all sDEGs from the integrated T2D and TNBC datasets and applied XGBoost ML model separately for each disease. Following that we have calculate the SHAP values to evaluate the importance of genes in separation of the disease and normal samples. Then we selected the top 20 contributing genes as showed in Fig. S2 from each disease-specific model and identified 9 SFGs (S100A9, CIRBP, ARL15, USP10, GCOM1, AASS, PLAUR, DHRS12, and PSMD1) by intersecting the gene-set obtain from the both disease model which may reflecting shared molecular features that may play a role in both T2D and TNBC. The evaluation of each fitted model on training dataset for gene prioritization is displayed as ROC curve, AUC value, and Confusion matrix in the (see Fig. 3). (A) ROC curve of T2D ML model. (B) Confusion matrix of T2D ML model. (C) ROC curve of TNBC ML model. (D) Confusion matrix of TNBC ML model Additionally, model performance on independent datasets demonstrated strong predictive ability, with an AUC of 0.96 for T2D and 0.89 for TNBC (see Fig. 4). These results indicate that the models can efficiently distinguish disease samples (T2D/TNBC) from controls and support the reliability of the identified SFGs in the analysis. Model performance assessment with independent test datasets, (A) TNBC, (B) T2D Determination of shared key genes (sKGs) through gene-set intersection The final set of sKGs was defined by intersecting the gene lists generated from both the network-based and machine learning–based approaches. This integrative filtering narrowed the candidates to four highly robust genes S100A9, CIRBP, USP10, and PSMD1 (see Fig. 5). These genes consistently ranked as central regulators in the PPI network and as top contributors in the ML models, suggesting that they may act as shared molecular drivers underlying the biological connection between T2D and TNBC. All downstream analyses were subsequently performed using these four high-confidence sKGs to further characterize their potential roles in the shared pathogenic landscape. sKGs Revealing the shared regulatory network and disease mechanisms of sKGs Identification of key regulator of sKGs We conducted regulatory network analyses of the sKGs using NetworkAnalyst to identify shared upstream regulators. A TFs-sKGs interaction network was constructed using the JASPAR database, revealing three highly connected TFs (TP53, USF1, and EGR1). Similarly, a miRNA-sKGs interaction network generated from TarBase and identified three key post-transcriptional regulators (hsa-miR-20a-5p, hsa-miR-18a-5p, and hsa-let-7b-5p). Together, these networks highlight how the sKGs are modulated at both transcriptional and post-transcriptional levels showed in Fig. 6, providing deeper insight into their upstream regulatory architecture (A) sKGs-TFs interaction network highlighting sKGs and crucial TFs. (B) sKGs-mi-RNAs interaction network highlighting sKGs and crucial miRNAs Revealing shared disease mechanisms mediated by sKGs We performed GO and KEGG pathway enrichment analyses for the four sKGs using Enrichr and GeneCodis4 to explore pathogenetic processes linking T2D with TNBC. For interpretation, we focused on the most significant terms across the three GO domains like biological process (BP), cellular component (CC), and molecular function (MF) as well as key KEGG pathways based on statistical significance and functional overlap. The most significant BP terms were autocrine signaling, chronic inflammatory response, and regulation of autophagy. The top CC terms included the nucleus, secretory granule lumen, and proteasome complex. The most significant MF terms were RAGE receptor binding, ubiquitin protein ligase binding, calcium-dependent protein binding, and Toll-like receptor binding. KEGG pathway analysis identified IL-17 signaling, proteasome, and pathways of neurodegeneration as the most enriched pathways. The results are displayed in the Table 3 below. These enrichment outputs provide a functional overview of the four sKGs from a bioinformatics and biological standpoint, supporting their relevance for downstream analyses. Table 3Important shared disease mechanisms of sKGs Annotation IDBPAgreed DatabasesAssociated sKGsGO:0035425autocrine signalingEnrichr, GeneCodis4S100A9GO:0002544chronic inflammatory responseEnrichr, GeneCodis4S100A10GO:0006914autophagyEnrichr, GeneCodis4S100A9, USP10GO:0048255mRNA stabilizationEnrichr, GeneCodis4CIRBPAnnotation IDCCAgreed DatabasesAssociated sKGsGO:0005634nucleusEnrichr, GeneCodis4S100A9, USP10, PSMD1, CIRBPGO:0034774secretory granule lumenEnrichr, GeneCodis4PSMD1, S100A9GO:0000502proteasome complexGeneCodis4PSMD1Annotation IDMFAgreed DatabasesAssociated sKGsGO:0035325Toll-like receptor bindingEnrichrS100A9GO:0005125Cytokine ActivityEnrichrS100A9GO:0050786RAGE Receptor BindingEnrichr, GeneCodis4S100A9GO:0031625ubiquitin protein ligase bindingEnrichr, GeneCodis4PSMD1GO:0048306calcium-dependent protein bindingGeneCodis4S100A9Annotation IDKEGGAgreed DatabasesAssociated sKGshsa04657IL-17 signaling pathwayEnrichr, GeneCodis4S100A9hsa03050ProteasomeEnrichr, GeneCodis4PSMD1hsa05022Pathways of neurodegenerationEnrichr, GeneCodis4PSMD1 Important shared disease mechanisms of sKGs sKGs-guided drug repurposing for therapeutic discovery Molecular docking for the repurposing of potential drug compounds was performed using AutoDock Vina. Prior to docking, we assessed the druggability of the sKGs and TFs and found that all proteins had druggability scores greater than 0.5, suggesting a strong potential for interaction with drugs (see Table S5). Then we collected the 3D structures of the four sKGs and three TFs from the PDB database and SWISS-MODEL. The experimentally resolved structures for S100A9, CIRBP, USP10, PSMD1, TP53, and EGR1 were obtained from the PDB database with the following IDs: 7UI5, 1X5S , 8TH6, 6CO4, 1H26, and 4R2A, respectively. The predicted 3D structure of USF1 was retrieved from SWISS-MODEL (model ID: 8IA3). The 3D structures of the selected drug candidates were retrieved from the PubChem database. All protein and ligand structures underwent preprocessing using Avogadro, Discovery Studio, and AutoDockTools4 to prepare them for docking. Binding affinity scores (BAS) between each protein–ligand pair were examined manually (see Table S6), and the top five compounds with the strongest predicted binding affinities (BAS < −7 kcal/mol) were selected for further analysis, as higher binding affinity generally indicates a greater likelihood of forming stable and biologically relevant interactions. The Top 20 interaction scores of proteins and drugs shown below in the heatmap Fig. 7. Binding affinity scores of top 20 drug molecules with sKGs Pharmacokinetics and toxicity analysis To further evaluate the repurposed drug candidates, we assessed their pharmacokinetic profiles and toxicity levels, as docking alone is not sufficient to determine whether a compound can perform effectively and safely in the human body. The five top-ranked compounds were examined using the DeepPK and pkCSM platforms, which predict ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) characteristics based on molecular structure. For each compound, we evaluated the major ADMET parameters in a stepwise manner. Absorption properties were examined through Caco-2 permeability, human intestinal absorption, and P-gp inhibition. Distribution was assessed by predicting blood–brain barrier penetration and CNS accessibility. Metabolic behavior was evaluated based on the inhibitory potential toward key cytochrome P450 enzymes, specifically CYP3A4 and CYP2D6. Excretion profiles were characterized using predicted total clearance and half-life values. Toxicity was examined through AMES mutagenicity, fathead minnow toxicity, and rat acute toxicity predictions. Together, these assessments showed in Table 4 provided a comprehensive overview of the pharmacokinetic and safety characteristics of the candidate compounds prior to further experimental evaluation. Table 4Pharmacokinetics analysis of top 5 drug compoundsCompoundsAbsorptionDistributionMetabolismExcretionToxicityCaco-2 (logPaap)HIAP-gp InhibitorBBBCNS (logPS)CYP3A4 InhibitorCYP2D6 inhibitorTC (log ml)Half-lifeAMES MutagenesisFathead Minnow −log10[(mg/L)/(1000*MW)]Rat (Acute) log[1/(mol/kg)]Tucatinib−5.21AbsorbedInhibitorNon-Penetrable−2.1InhibitorNon-Inhibitor5.43<3hsSafe11.062.99Pimozide−5.29AbsorbedInhibitorPenetrable−1.97InhibitorInhibitor8.66<3hsSafe6.452.91Tepotinib−5.18AbsorbedInhibitorPenetrable−2.62Non-InhibitorNon-Inhibitor5.96<3hsSafe12.682.95Ursodiol−5.26AbsorbedInhibitorPenetrable−2.4Non-InhibitorNon-Inhibitor5.21<3hsSafe3.892.3Zanubrutinib−5.93AbsorbedInhibitorPenetrable−1.98InhibitorNon-Inhibitor6.99<3hsSafe6.692.67 Pharmacokinetics analysis of top 5 drug compounds All selected compounds showed favorable absorption profiles, reflected by consistent Caco-2 permeability values and positive predictions for human intestinal absorption. In addition, all compounds were predicted to inhibit P-gp, suggesting a potential influence on intestinal drug transport. Regarding distribution, most compounds were predicted to be capable of crossing the blood–brain barrier, whereas Tucatinib was classified as non-penetrable. The CNS permeability values (logPS) indicated moderate central nervous system accessibility across the compound set. Metabolic profiling revealed differences in cytochrome P450 interactions, with Tucatinib, Pimozide, and Zanubrutinib predicted to inhibit CYP3A4, while Pimozide also showed inhibitory potential toward CYP2D6. By contrast, tepotinib and ursodiol were predicted to have no inhibitory effects on either enzyme. All compounds displayed comparable total clearance values and short predicted half-lives of less than three hours. Toxicity predictions further indicated that none of the compounds were AMES mutagenic, and both fathead minnow and rat acute toxicity estimates remained within acceptable range. Drug likeliness property analysis We evaluated their drug-likeness properties using the Lipinski Rule of Five using DeepPK and pkCSM platforms to understand therapeutic potential of the candidate potential drugs. The physicochemical characteristics of the selected compounds are presented in Table 5. Overall, the compounds showed comparable molecular weights, remaining within a range typical of small-molecule drugs. Differences in lipophilicity were observed, with pimozide and tucatinib displaying relatively higher LogP values, whereas tepotinib, ursodiol, and zanubrutinib exhibited more moderate lipophilicity. Molecular flexibility, as reflected by the number of rotatable bonds, ranged from four to seven across the compounds, suggesting moderate conformational adaptability. Variations were also noted in hydrogen bonding capacity, with tucatinib and tepotinib possessing higher numbers of hydrogen bond acceptors, while ursodiol showed a greater number of hydrogen bond donors. In addition, the calculated surface area varied considerably among the compounds, highlighting structural diversity that may influence molecular interactions and pharmacokinetic behavior. Table 5Drug likeliness properties of the selected five compoundsDescriptorTucatinibPimozideTepotinibUrsodiolZanubrutinibMolecular Weight480.532461.556492.583392.58471.561LogP5.0945.8574.0084.4784.2226Rotatable Bonds67746H. Acceptors104846H. Donors21032Surface Area206.59841.03215.36677.76203.917 Drug likeliness properties of the selected five compounds Validation of the potential drug molecules Among the five candidate compounds, tepotinib and ursodiol showed the most favorable pharmacokinetic and drug-likeness profiles. Notably, these two compounds were predicted to be non-inhibitors of both CYP3A4 and CYP2D6, suggesting a reduced risk of potential drug–drug interactions. In addition, tepotinib and ursodiol did not violate any Lipinski’s rule-of-five criteria, indicating more suitable drug-like characteristics compared with the other candidates. We further validated the selected drug candidates to assess their broader therapeutic applicability in both diseases. For this purpose, the top 10 key genes (KGs) were identified separately for T2D and TNBC, and molecular docking was performed against these targets. Almost all the BAS is less than −7 kcal/mol, as shown in the Fig. 8, indicated that most target proteins exhibited strong binding interactions with the selected compounds. These results suggest that the candidate drugs are capable of interacting effectively with disease-specific targets in both conditions. This validation step supports the suitability of the two prioritized drug compounds for further experimental evaluation. Binding affinity scores (BAS) of two potential drug molecules with key genes of both disease (A) BAS with TNBC (B) BAS with T2D Molecular dynamics analysis Following molecular docking, the three highest-scoring protein–ligand complexes were selected for 100 ns molecular dynamics (MD) simulations to evaluate their structural stability and interaction dynamics under simulated physiological conditions. System behavior during the simulations was assessed using key parameters, including root mean square deviation (RMSD), root mean square fluctuation (RMSF), solvent-accessible surface area (SASA), radius of gyration (Rg), and MM-PBSA binding free energy calculations. Prior this, we examined the protein-drug binding positions with our selected potential drug compound with 2D schematic diagrams of receptor–ligand interactions and 3D views of the complexes provided in Fig. 9, highlighting interacting residues. Top-ranked two drug-target complexes highlighting their 3-dimension (3D) view and interacting residues RMSDBb analysis was employed to monitor the temporal backbone stability of each protein–ligand complex by measuring deviations from the initial structure over the course of the simulation The RMSDBb plot shows that both complexes reach equilibrium early in the simulation and remain stable afterward. However, the CIRBP–ursodiol complex consistently exhibits lower RMSDBb values than the CIRBP–tepotinib complex throughout the simulation showed Fig. 10(A), indicating that ursodiol binding induces smaller backbone deviations than tepotinib. Molecular dynamics analysis of two top ranked CIRBP complexes with tepotinib (red) and ursodiol (green). (A) backbone RMSD, (B) Radius of gyration, (C) solvent-accessible surface area, (D) residue-wise RMSF, and (E) frame-wise MM-PBSA binding free energy over 100 ns simulation time The radius of gyration (Rg) reflects the compactness of the protein structure. The Ursodiol-bound complex maintains a lower and more stable Rg showed in Fig. 10(B), indicating a more compact and tightly folded structure than tepotinib-bound complex which showed which higher Rg with larger early fluctuations, suggesting a relatively looser packing before stabilization. Solvent-Accessible Surface Area (SASA) decreases over time in both systems showed in Fig. 10(C), indicating gradual structural tightening during the simulation. Root mean square fluctuation (RMSF) was used to understand the residue-level local flexibility showed in Fig. 10(D), both complexes displayed higher fluctuations at the N- and C-termini, which is expected due to intrinsic terminal flexibility. Importantly, most internal residues showed lower RMSF values, indicating reduced local flexibility and better stabilization of the protein backbone. The binding free energy plot in the Fig.10(E) showed time-dependent MM-PBSA values for both complexes over the simulation time. While both systems display fluctuations, the average binding energies remain within a comparable range, with tepotinib showing slightly more favorable energetic contributions at certain intervals. Although Ursodiol exhibited superior structural stability than tepotinib, MM-PBSA energies showed overlapping distributions, highlighting that structural stability and binding energetics are not always directly correlated. Overall, the results demonstrate that ursodiol offers superior structural stabilization, whereas tepotinib contributes relatively stronger energetic interactions. Discussion Type 2 diabetes (T2D) is a metabolic disorder highly recognized as a key risk factor for triple negative breast cancer (TNBC), as chronic hyperglycemia, insulin resistance, and systemic inflammation can create a tumor-promoting environment and accelerate cancer progression [10]. The management of TNBC with T2D as comorbidity may be complicated than the management of either disease alone without comorbidity, since proper treatment strategies during their co-occurrence/existence is not yet investigated rigorously. This study detected shared key-genes (sKGs) and their mechanism associated with both diseases for disclosing a tentative treatment strategy during their co-occurrence/existence. Transcriptomic analyses have emerged as a robust approach for investigating a wide range of diseases, including autoimmune disorders, viral infections, cancers, and neurodegenerative conditions [65]. In addition to transcriptomics-based biomarker discovery, recent studies have also explored biomarker identification at the genetic variant level using integrative statistical and pathway-based approaches. For example, a recent study demonstrated that combining statistical analysis with pathway enrichment can identify variant-level biomarkers associated with aggressive cancer phenotypes and provide deeper insight into disease mechanisms [66]. Through transcriptomic analyses, at first, we identified 36 shared DEGs (sDEGs) between T2D and TNBC by combining five transcriptomics profiles GSE25724, GSE9006, GSE29221, GSE64998, and GSE29231 for T2D and five transcriptomics profiles GSE27447, GSE38959, GSE37751, GSE43358, and GSE65194 for TNBC. Then we identified 4 sKGs (S100A9, CIRBP, USP10, and PSMD1) that are significantly enriched in crucial biological processes (GO terms) and signaling pathways (KEGG), highlighting their potential involvement in the comorbid mechanisms underlying these two diseases via machine learning and topological based analysis. The link of sKGs with T2D and TNBC are also supported by the literature review of individual T2D and TNBC studies as discussed below. Among our identified sKGs, S100A9 is critical for chronic inflammatory response and autocrine signaling, activating TLR4 and RAGE pathways that worsen insulin resistance and sustain low-grade inflammation in T2D [67–69]. This inflammatory milieu may create a permissive environment for tumor initiation and progression, as S100A9 is closely associated with the IL-17 signaling pathway, a major driver of tumor-promoting inflammation [69, 70]. This IL-17 signaling pathway plays a key role in inflammation, insulin resistance, and the development of T2D, and it is also strongly involved in breast cancer proliferation [71, 72]. At the molecular level, S100A9 exhibits Toll-like receptor binding, RAGE receptor binding, cytokine activity, and calcium-dependent protein binding, enabling amplification of inflammatory signals shared by both metabolic and oncogenic processes [73–75]. In TNBC, elevated S100A9 in tumor and myeloid cells promotes invasion, epithelial–mesenchymal transition, and aggressive tumor behavior [76, 77]. Notably, S100A9-mediated dysregulation of autophagy may allow cancer cells to adapt to the metabolic stress imposed by diabetes [78]. S100A9’s role may be critical in coordinating chronic inflammatory signaling and metabolic stress responses, both of which are dysregulated in T2D and collectively contribute to an immune-permissive tumor microenvironment that accelerates TNBC progression. PSMD1, a regulatory subunit of the 26S proteasome, may contribute to the shared pathological mechanisms between T2D and TNBC through its involvement in the Proteasome, which governs protein turnover and cellular stress adaptation [78–80]. Its role in ubiquitin-dependent protein degradation and ubiquitin ligase interactions suggests that PSMD1 may influence the stability of key metabolic and cell-cycle regulators under disease conditions [81, 82]. Localization within the proteasome complex, nucleus, and secretory granule lumen may enable PSMD1 to coordinate transcriptional control and stress-responsive signaling networks [82, 83]. In TNBC, elevated PSMD1 expression may support tumor progression by reinforcing proteasome dependency and facilitating the degradation of tumor-suppressive proteins, a hallmark of aggressive BC phenotypes [84]. Meanwhile, in T2D, dysregulated proteasome activity may promote insulin resistance and chronic inflammation, indicating that altered PSMD1-mediated proteostasis could represent a molecular link connecting metabolic dysfunction with TNBC progression. CIRBP, a stress-responsive RNA-binding protein, may contribute to shared mechanisms between T2D and TNBC by regulating mRNA stabilization during metabolic and inflammatory stress conditions [84, 85]. Through its ubiquitin protein ligase binding activity and predominant localization in the nucleus, CIRBP may influence transcript turnover, genome stability, and stress-adaptive gene expression programs relevant to both metabolic dysregulation and tumor progression [85, 86]. Emerging evidence suggests that CIRBP overexpression is associated with poor prognosis and enhanced aggressiveness in TNBC, potentially by stabilizing oncogenic and survival-related transcripts under stress conditions [87, 88]. In parallel, chronic low-grade inflammation and cellular stress in T2D may alter CIRBP-mediated RNA regulation, thereby amplifying inflammatory signaling and metabolic imbalance [89, 90]. Collectively, these findings suggest that CIRBP-driven RNA-stability mechanisms and stress-response pathways linked to the stress–response–related signaling network may represent a molecular bridge connecting metabolic dysfunction with TNBC progression. USP10, a deubiquitinating enzyme, may link T2D and TNBC by regulating protein stability and cellular stress responses through autophagy [91, 92]. Its activity in the nucleus allows USP10 to influence transcriptional programs and stress-adaptive signaling that are disrupted in metabolic disease and cancer progression [93, 94]. In TNBC, elevated USP10 promotes tumor growth and metastasis by stabilizing oncogenic proteins and supporting survival under stress conditions [91, 93]. Meanwhile, USP10-mediated regulation of autophagy and metabolic signaling may contribute to insulin resistance and chronic inflammation observed in T2D [95, 96]. Together, these findings suggest that dysregulated USP10 activity may act as a molecular bridge connecting metabolic imbalance with aggressive TNBC behavior. All sKGs are regulated by transcription factors TP53, USF1, and EGR1. TP53, beyond its classical tumor suppressor role, is increasingly recognized as a central regulator of cellular metabolism, oxidative stress, and insulin sensitivity [97]. Under diabetic conditions, chronic hyperglycemia and oxidative stress can impair TP53 function, leading to reduced apoptosis and increased survival of damaged cells, thereby facilitating tumor initiation and progression [98, 99]. TP53, the well-known tumor suppressor, regulates apoptosis, DNA repair, and metabolic pathways, and its impairment under diabetic conditions can enhance genomic instability and cancer risk [95, 100, 101]. USF1 is a critical transcriptional regulator of lipid and glucose homeostasis. In T2D, USF1 dysregulation contributes to insulin resistance, dyslipidemia, and altered energy metabolism [102]. These metabolic disturbances can create a nutrient-rich, growth-promoting microenvironment that supports tumor proliferation and survival in TNBC [103, 104]. Moreover, USF1 has been implicated in regulating genes involved in cell cycle progression and metabolic reprogramming, suggesting that it may act as a molecular bridge linking metabolic dysfunction to cancer cell adaptation [105, 106]. EGR1 is an immediate early response transcription factor that is rapidly induced by stress signals, including inflammation, hypoxia, and metabolic imbalance [107]. In T2D, persistent low-grade inflammation can lead to sustained activation of EGR1, which in turn regulates genes involved in cytokine production, fibrosis, and cellular stress responses [108, 109]. In TNBC, EGR1 can exhibit dual roles depending on the cellular context, but chronic activation is often associated with enhanced proliferation, angiogenesis, and tumor progression [108]. At the post-transcriptional level, the identified miRNAs: hsa-miR-20a-5p, hsa-miR-18a-5p, and hsa-let-7b-5p further refine gene expression programs and contribute to disease progression. MiR-20a-5p and miR-18a-5p, members of the miR-17–92 cluster, are well-known oncogenic miRNAs that promote cell proliferation, inhibit apoptosis, and facilitate cell cycle progression in TNBC [110, 111]. In the context of T2D, these miRNAs have also been linked to impaired insulin signaling, increased inflammatory responses, and altered glucose metabolism [112, 113]. Their upregulation may therefore amplify both metabolic dysfunction and tumorigenic processes simultaneously [111]. In contrast, let‑7b‑5p generally functions as a tumor suppressor by inhibiting oncogenes and regulating cell differentiation and metabolic pathways [114]. However, under conditions of chronic metabolic stress, such as those present in T2D, the expression or activity of let‑7b‑5p may be reduced [115]. This reduction can relieve repression of oncogenic targets, thereby promoting tumor progression in TNBC while also contributing to metabolic dysregulation [110]. Dysregulation of these TFs and miRNAs can orchestrate a feed-forward loop enhancing inflammation, epithelial-mesenchymal transition, and metabolic dysfunction, may creating a mechanistic axis that bridges T2D and TNBC progression. To identify potential therapeutic drug candidates against TNBC associated with T2D, we focused on the sKGs and crucial sTFs identified and validated in this study. For drug repurposing, disease-associated drug molecules were collected from published literature and the DrugBank database, while sKG-guided drugs were obtained from the DGIdb database. Through molecular docking we have identified 5(Tucatinib, Pimozide, Tepotinib, Ursodiol, and Zanubrutinib) top ranked drug molecules which have showed highest interaction score. We then assessed their ADME/T characteristics and drug-likeness properties to further refine the list. Based on this combined evaluation, tepotinib and ursodiol (Ursodeoxycholic acid) emerged as the most promising candidates for further investigation in this complex disease context. Following this we also assessed how these two candidate drug compounds have interacted with the highest ranked protein CIRBP in molecular docking analysis and found that both compounds were observed to occupy the predicted binding pocket of CIRBP and form multiple non-covalent interactions that contribute to complex stability. In the case of tepotinib, the interaction network was characterized by a combination of hydrogen bonding and hydrophobic contacts involving key residues such as PHE16, ALA89, LYS91, GLN88, and ARG85. These interactions suggest a well-anchored ligand orientation within the binding site, supported by π–π stacking and electrostatic interactions that enhance binding specificity. Similarly, ursodiol demonstrated slightly stronger binding affinity and formed extensive interactions with residues including PHE16, PHE58, VAL45, LYS91, and SER92, indicating a stable and complementary fit within the CIRBP binding pocket. The presence of both polar and hydrophobic interactions suggests that ursodiol may achieve a balanced interaction profile, contributing to its binding stability. Overall, the consistent involvement of key residues such as PHE16 and LYS91 across both complexes highlights their potential importance in ligand recognition and binding. These findings suggest that both tepotinib and ursodiol can effectively interact with CIRBP at the molecular level, supporting their potential as candidate compounds for further investigation. The molecular dynamics (MD) simulation analyses, including SASA, RMSD, RMSF, and radius of gyration over a 100 ns trajectory, indicated stable interactions between CIRBP and both tepotinib and ursodiol. These results further support the stability of the protein-ligand complexes and reinforce the potential of these compounds as promising candidates. Both treatment are FDA-approved drug for treating diseases [116, 117]. Recent studies have shown that tepotinib can act as an effective inhibitor of α-glucosidase activity, suggesting its potential to help control hyperglycemia, a key metabolic feature of T2D [118]. Tepotinib has demonstrated the ability to suppress tumor cell proliferation, invasion, and migration while inducing apoptosis through inhibition of the MET and PI3K/AKT signaling pathways, which are established therapeutic targets in several malignancies, including TNBC and melanoma [119]. Ursodiol is an FDA-approved therapeutic drug that exhibit pro-apoptotic effects in BC cell lines, including those relevant to TNBC [120]. Administration of ursodiol has been associated with improvements in anthropometric measures, liver function parameters, and diastolic blood pressure in patients with T2D [121]. It has been reported that Ursodiol enhance the therapeutic effects of the dipeptidyl peptidase-4 inhibitor sitagliptin by increasing glucagon-like peptide-1 secretion in patients with T2D and chronic liver disease [122]. The mechanistic profiles of these drugs suggest that; they may have the potential as therapeutic options for TNBC in patients with T2D. However, although in silico approaches including molecular docking, ADMET profiling, and toxicity prediction offer valuable preliminary insights, comprehensive experimental validation is required before progression to clinical trials. Conclusion This study identified top-ranked four shared key genes (sKGs): S100A9, CIRBP, USP10, and PSMD1 associated with both T2D and TNBC through the protein-protein interaction (PPI) network and explainable machine learning approaches from the 36 shared differentially expressed genes (sDEGs) obtain by merging 5 transcriptomics datasets for each of T2D and TNBC. The sKGs enrichment analysis with GO-terms and KEGG pathways disclosed some crucial biological processes, cellular components, molecular function and pathways that might be linked to the development of both diseases. For instances, the enriched signaling pathway ‘IL-17’ may play crucial role in stimulating both diseases. sKGs versus TFs interaction analysis discovered top-ranked 3 proteins (TP53, USF1, and EGR1) as the key transcriptional regulators of sKGs, where TP53 and USF1 might be played the significant role in occurring both diseases. The interaction between sKGs and miRNAs revealed top-ranked 3 miRNAs (hsa-miR-20a-5p, hsa-miR-18a-5p, and hsa-let-7b-5p) as the post transcriptional regulators. Finally, sKGs-guided two candidate drug molecules Tepotinib and Ursodiol were suggested by using molecular docking, ADME/T, drug likeliness analysis and MD simulation analysis. Thus, the findings might be useful resources for taking a treatment plan against TNBC with T2D as comorbidity. Limitation of this study In this study, we analyzed independent transcriptomic datasets for TNBC and T2D to investigate both diseases associated sKGs. This approach was adopted because no publicly available transcriptomic datasets derived from patients who are simultaneously affected by both diseases during this study. Consequently, molecular docking-based drug screening by targeting sKGs may also identify certain false-positive or false-negative candidate molecules. Therefore, the findings require experimental (in-vitro and in-vivo) and clinical validation to develop a proper treatment strategy for TNBC with T2D as comorbidity. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Author contribution M.F.A.: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Visualization, Writing - original draft. O.S.: Formal analysis, Methodology, and visualization, M.A.L.: Investigation, Visualization, M.F.A.: Data curation, Formal analysis. M.B.: Data curation, Visualization, D.M.A.: Data curation, Visualization, M.S.H.: Investigation, Visualization, M.N.H.M.: Conceptualization, Methodology, Software, Supervision, and Writing - review & editing. Funding This study was conducted without any funding. Data availability No datasets were generated or analysed during the current study. Declarations Ethics approval and consent to participate Not applicable, since the datasets analyzed in this study were downloaded from online sources that are freely available without any restriction. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement No datasets were generated or analysed during the current study.