Exploring the toxicological impact of DEHP exposure on colorectal cancer through network toxicology, machine learning and bioinformatics analysis Authors: Ling Wang, Yuning Qin, Wenbin Fan Published: 2025 ABSTRACT ObjectiveColorectal cancer (CRC) ranks among the most prevalent malignant tumors, yet its underlying mechanisms remain not fully understood. The role of environmental factors is critical in its development and progression. Di-(2-ethylhexyl) phthalate (DEHP), a widespread environmental contaminant, poses significant hazards to human health. Prior research indicates that exposure to DEHP can disrupt cellular processes like proliferation, differentiation, and apoptosis, which may contribute to cancer development. Nevertheless, the specific role and molecular pathways involved with DEHP in CRC are still poorly defined.MethodsWe integrated multiple public databases to identify overlapping targets of DEHP and CRC. Machine learning methods were applied to prioritize core targets, whose expression levels and diagnostic performance were then validated in the Gene Expression Omnibus (GEO) datasets. The association between core targets and immune cell infiltration was evaluated using CIBERSORT-based deconvolution. Finally, molecular docking was performed to predict the interactions between DEHP and the hub proteins, and, based on the docking results, an explicit-solvent molecular dynamics (MD) simulation of the top-ranked DEHP–target complex was conducted to further assess binding stability.ResultsAnalysis of publicly available databases revealed 57 potential targets that were significantly enriched in signaling pathways related to “microRNAs in cancer” and “chemical carcinogenesis–receptor activation.” Machine learning methods identified five primary targets: CASP3, BCL6, BRD4, PPARA, and PRKCD. The expression levels of these genes in CRC tissues were significantly different from those in control tissues and showed reasonable diagnostic performance, as well as correlations with immune cell infiltration. Molecular docking suggested that DEHP can bind all five hub proteins, with PPARA exhibiting the most favorable binding affinity (− 6.7 kcal/mol). The 100-ns MD simulation further supported this binding mode by demonstrating that the DEHP–PPARA complex maintains a dynamically stable conformation with persistent key interactions.ConclusionsOur findings generate preliminary mechanistic hypotheses on how DEHP exposure may contribute to CRC by perturbing known cancer-related pathways and reshaping the intestinal immune microenvironment, with PPARA emerging as a key candidate mediator. The study highlights the utility of combining network toxicology, machine learning, immune deconvolution, molecular docking, and MD simulations to evaluate the potential carcinogenic risks of environmental pollutants.Supplementary InformationThe online version contains supplementary material available at 10.1186/s40360-025-01065-0. Introduction Colorectal cancer (CRC) is a prevalent type of malignant tumor, holding the second position for incidence and the fourth for mortality among all malignant tumors, respectively [1]. According to the latest statistical data, the incidence rate of CRC ranks among the top in malignant tumors, and its mortality rate is also relatively high [2]. The pathogenesis of CRC has not yet been fully elucidated and involves multiple factors, including genetic and environmental influences [3]. Although genetic or gene mutations may account for 10% to 15% of CRC cases, the risk attributed to environmental factors is as high as 80% to 90% [4]. Therefore, CRC is a disease influenced by environmental factors. Di-(2-ethylhexyl) phthalate (DEHP) serves as a common plasticizer extensively utilized in everyday items. This compound is present in various household products, medical instruments, rubbing alcohol, liquid soaps, and packaging for food. Additionally, it can be found in environmental samples such as food, air contaminants, industrial wastewater, soil, and waterways [5].Due to its high additive rate and non-covalent bonding characteristics, DEHP is highly prone to being released into environmental media during both usage and disposal stages [6]. Research has shown that DEHP can cause diseases in multiple organs, including the thyroid, testes, uterus, ovaries, liver, and nerves [7, 8]. In addition, contact with DEHP could disrupt hormonal equilibrium, resulting in the misregulation of cellular activities like cell growth, differentiation, and programmed cell death, which may contribute to cancer development [9, 10]. Recently, Chen et al. found that DEHP can enhance resistance to chemotherapy in CRC cells by altering metabolic pathways, which diminishes the effectiveness of chemotherapeutic agents [11]. Su et al. have demonstrated that exposure to DEHP can lead to alterations in the glycosylation of cell surfaces in patients with CRC, which enhances the traits of cancer stem cells and consequently accelerates tumor development [12]. Considering that the toxic effects of DEHP require further exploration and that related risk prediction and exposure models need improvement, it is necessary to adopt a network toxicology strategy for risk assessment and toxicity prediction of environmental pollutants. The combination of network toxicology and molecular docking represents a highly effective and precise method for assessing drug side effects and toxicity [13]. Network toxicology examines how chemical substances adversely affect organisms through bioinformatics techniques and computer modeling [14]. This approach can forecast and assess the toxic impacts of chemicals on humans by analyzing the interaction networks between chemical substances and biological molecules [15], thereby aiding in the comprehension of the toxicological mechanisms of these substances and facilitating the identification of potential toxins. Molecular docking serves to model the interactions between target molecules and biomolecules, allowing predictions regarding their affinity and potential biological effect [16].Furthermore, the toxicity of various compounds can be evaluated through the analysis of their interactions and binding affinities with specific targets. A lower binding energy between ligands and receptors indicates a more stable ligand-receptor complex and a higher probability of interaction [17]. Given the widespread human exposure to DEHP and the established role of inflammation and immune dysregulation in CRC, there is a need to clarify how this plasticizer may intersect with CRC-related molecular pathways. In this study, we adopted an integrated network toxicology and bioinformatics framework to address this gap. Rather than searching for entirely new oncogenic drivers, we sought to re-interpret well-known CRC-related genes as potential mediators of DEHP toxicity. To identify core genes, we first intersected a comprehensive set of DEHP-related toxicological targets with genes associated with CRC and then applied three complementary machine-learning algorithms. These genes were then validated in independent GEO cohorts, analyzed in relation to immune-cell infiltration, and further examined by molecular docking and molecular dynamics simulations. This environmental pollutant–centered and multi-layer integrative approach aims to generate mechanistic hypotheses on how DEHP may contribute to CRC development through remodeling of the intestinal immune microenvironment. Methods Toxicity analysis and prediction In this research, we employed two online tools, ProTox 3.0 (https://tox.charite.de/; accessed on 4 July 2025), which serves as a server for predicting chemical toxicity, and ADMETlab 2.0 (https://admetmesh.scbdd.com; accessed on 4 July 2025), a thorough platform for forecasting pharmacokinetics and toxicity, to evaluate the toxicological profile of DEHP through its SMILES notation. The detailed predictions regarding various toxicological features and pharmacokinetic attributes of DEHP offered by these resources were instrumental in assessing its safety and associated risks. Target construction of DEHP The standard structure and SMILE node details of DEHP were established by conducting a search for “Di-(2-ethylhexyl) phthalate” within the PubChem database (https://pubchem.ncbi.nlm.nih.gov/; accessed on 4 July 2025). From the search outcomes, potential targets for DEHP were extracted from the ChEMBL database (https://www.ebi.ac.uk/chembl/; accessed on 4 July 2025) using the term “Di-(2-ethylhexyl) phthalate,” while restricting the search to “Homo sapiens” species. Following this, the SMILE notation for DEHP was submitted to the STITCH database (http://stitch.embl.de/; accessed on 4 July 2025) to discover any potential targets that might have been missed. The structures from the search results were subsequently examined for their consistency. The Swiss Target Prediction online tool (http://www.swisstargetprediction.ch/; accessed on 4 July 2025) was employed to forecast possible action targets for DEHP, using “Probability ≥ 0” as the screening criterion. To avoid missing potentially relevant DEHP-responsive targets, we deliberately adopted an inclusive screening strategy at this initial stage and collected all non-zero hits from these toxicology-oriented resources. Finally, the potential targets from these databases were integrated and deduplicated, and these curated targets were utilized to develop the target library for DEHP. Target construction of colorectal cancer After conducting a thorough review of both national and international scholarly articles, along with the Genecards database (https://www.genecards.org/; accessed on July 7, 2025), the Therapeutic Target Database (http://db.idrblab.net/ttd/; accessed on July 7, 2025), and the OMIM database (https://www.omim.org/; accessed on July 7, 2025), we identified instances of CRC using the search term “colorectal cancer,” noting the data after eliminating duplicates. Following this, we utilized R software to filter the information and generate a Venn diagram, where the overlapping area was regarded as potential toxicological targets of DEHP related to CRC. Functional enrichment analysis based on GO and KEGG pathways KEGG and GO enrichment analyses of the intersecting genes were performed using the “clusterProfiler”, “enrichplot”,“org.Hs.eg.db”, and “ggplot2” R packages. Core target selection by machine learning algorithms To pinpoint essential targets linked to DEHP-triggered CRC, three distinct machine learning prediction models were employed: the Random Forest (RF) model, the Support Vector Machine Recursive Feature Elimination (SVM-RFE) model, and the Least Absolute Shrinkage and Selection Operator (LASSO) regression model. The SVM-RFE approach was executed utilizing the “e1071” and “caret” R packages. Key genes were identified through the process of feature elimination. We then classified the significant genes with the “randomForest” package in R. A regression model was developed using the “glmnet” package in R, with the model designated as “binomial,” and the optimal lambda (λ) value was chosen based on “lambda.min.” A log-lambda graph depicting the LASSO coefficients was generated. Following the screening process with all three algorithms, the outcomes from each method were compared, and the genes that were found in common were recognized as potential core targets for CRC toxicity induced by DEHP. Differential expression of core genes and assessing diagnostic accuracy We obtained the GEO datasets GSE21510 and GSE44076, which provide gene-expression profiles from colorectal tumor tissues and corresponding non-tumor intestinal mucosa, and used them to analyze and visualize the expression patterns of the essential genes. These datasets offer sufficient sample sizes, high-quality clinical annotation, and Affymetrix-based platforms, and have been widely used as validation cohorts in CRC transcriptomic studies [18, 19]. The GEO database was employed to download the expression matrices and sample-level metadata. Background correction and quantile normalization were performed using the limma package. Expression data were normalized using the normalizeBetweenArrays function from the limma package, followed by log2 transformation. Probe IDs were converted to official gene symbols according to the platform annotation files (GPL570 and GPL13667), and when multiple probes mapped to the same gene, their the mean expression value was used. For analyses requiring integration of the two cohorts, potential batch effects were corrected using the ComBat function in the sva package, ensuring that batch differences were reduced while biological differences between tumor and normal tissues were preserved. Genes with a p-value < 0.05 were deemed significantly differentially expressed. Using mRNA expression data from normal and CRC samples in GSE21510 and GSE44076, we further constructed receiver operating characteristic (ROC) curves to assess the diagnostic relevance of the identified core targets, and calculated the area under the curve (AUC) to evaluate their diagnostic performance. Immune infiltration analysis The correlation analysis of infiltrating immune cells was carried out utilizing the CIBERSORT algorithm, which measured the relative expression levels of 22 different types of immune cells in both patients with CRC and healthy individuals. The analysis focused on the differences in immune cell populations between these two groups, as well as the relationships among various immune cell types. For visualization, the “corrplot” package in R software was used. Molecular docking The molecular docking method was employed to perform a thorough analysis of the intermolecular interactions involving DEHP and the key targets identified in this study by forecasting their binding modes and affinities. Crystal structures of these key targets were retrieved from the PDB database (https://www.rcsb.org/; accessed on 8 July 2025). With the assistance of PyMOL software (version 3.0.3), water molecules and the initial ligands were removed from the target proteins. Subsequently, these proteins were imported into AutoDock Tools software (version 1.5.7) for dehydration, hydrogenation, charge calculation, and the formation of non-polar hydrogen bonds. After defining the grid box dimensions and genetic algorithm parameters, molecular docking was carried out using AutoDock Vina software (version 1.2.3) with CMD command inputs. Finally, the results were visualized using Discovery Studio software (version 2019) and PyMOL, which enabled an interactive review of the predicted binding configurations in three-dimensional space, thereby aiding in the interpretation and evaluation of computational predictions. Molecular dynamics Following molecular docking, molecular dynamics (MD) simulations were conducted to further investigate, analyze, and evaluate the binding stability of the receptor-ligand complex. MD simulations more accurately replicate the biological macroenvironment, accounting for the influences of temperature, pressure, and charge on the binding interactions between the receptor and ligand [20]. In this study, GROMACS 2022 was utilized for molecular dynamics simulations. The force field parameters were generated using the pdb2gmx tool within GROMACS. The ligand topology was created using the sobtop_1.0(dev3.1) software with the GAFF2 force field, while charges were assigned via the RESP method. The receptor protein employed the AMBER14SB force field. For the simulation system, solvation was performed using the TIP3P water model within a cubic box of 1 nm. To maintain electroneutrality, Na+ and Cl− ions were added at a concentration of 0.15 M NaCl. Long-range electrostatic interactions were calculated using the PME method with a cutoff distance of 1 nm, and bond constraints were applied using the LINCS algorithm. Prior to the simulation, the system underwent a three-stage energy minimization process to sequentially optimize the solute, water molecules, and ions, ultimately refining the entire system. Under simulation conditions, the temperature was maintained at 310 K using the Nosé–Hoover thermostat, while the pressure was kept at 1 bar using the Parrinello–Rahman barostat. The simulation duration was set to 100 ns under the NPT ensemble with a time step of 2 fs. During the simulation, GROMACS tools were employed to calculate Root mean square deviation (RMSD), Root mean square fluctuation (RMSF), hydrogen bonds, radius of gyration (Rg), solvent-accessible surface area (SASA), and Free Energy Landscape (FEL) to analyze system stability and structural changes. Results ProTox and ADMETlab-prediction of toxicity of chemicals The predictive toxicological analysis of DEHP via the ProTox platform indicates that this substance exhibits notable carcinogenic potential, associated with nephrotoxicity and effects on the blood-brain barrier, implying a certain degree of carcinogenic risk (Fig. 1). Furthermore, the ADMEtlab platform highlights that DEHP significantly influences skin sensitization, carcinogenicity, and eye irritation. In summary, DEHP poses substantial carcinogenic threats, leading us to conduct a network toxicology evaluation for a more comprehensive understanding of the potentially increased risk of colorectal cancer (CRC) linked to DEHP. Both ProTox and ADMETlab predictions regarding the toxicity of various chemicals were utilized for additional insights. Fig. 1Toxicity distribution Map of DEHP Toxicity distribution Map of DEHP Potential targets of DEHP targets in colorectal cancer In this research, a total of 278 DEHP targets were initially identified from the ChEMBL, Swiss Target Prediction, and STITCH databases. Additionally, 2401 targets strongly linked to CRC were collectively recognized using the Genecards, TTD, and OMIM databases. By integrating and eliminating duplicates from these target collections, 57 common targets emerged, which were regarded as potential targets for CRC induced by DEHP (Fig. 2). Fig. 2The overlapped targets of DEHP and Colorectal cancer The overlapped targets of DEHP and Colorectal cancer KEGG and GO analysis of DEHP targets in colorectal cancer Utilizing R software to conduct KEGG enrichment analysis on the intersecting genes revealed that these genes were predominantly associated with several significant pathways. Notably, MicroRNAs in cancer emerged as the most prominent pathway, accompanied by relevant pathways such as Transcriptional misregulation in cancer and Chemical carcinogenesis-receptor activation, as illustrated in Fig. 3A. Furthermore, Gene Ontology (GO) analysis indicated that biological processes (BP) predominantly focused on the epigenetic regulation of gene expression. When examining cellular components (CC), the genes showed enrichment in histone methyltransferase complexes and methyltransferase complexes. In terms of molecular functions (MF), the analysis revealed a strong emphasis on functionalities such as DNA-binding transcription factor binding, RNA polymerase II-specific DNA-binding transcription factor binding, and transcription coregulator activity. These functions play critical roles in the regulation of gene transcription and other related processes(Fig. 3B). Consequently, the induction of CRC by DEHP appears to be a complex interplay of multiple pathways and regulatory mechanisms. Fig. 3Enrichment Analysis. KEGG Analysis (A), GO Analysis (B). In Fig. A and B, the vertical axis represents the enrichment items, and the horizontal axis represents the number of gene enrichments Enrichment Analysis. KEGG Analysis (A), GO Analysis (B). In Fig. A and B, the vertical axis represents the enrichment items, and the horizontal axis represents the number of gene enrichments Identification of key targets using machine learning Subsequently, we employed the LASSO, RF, and SVM algorithms to determine the key target genes among the 57 CRC-related targets influenced by DEHP. Initially, LASSO logistic regression was performed with a five-fold cross-validation process to fine-tune the penalty parameters, leading to the discovery of 15 genes that could serve as potential core targets (Fig. 4A and B). The SVM-RFE regression method also selected 15 genes (Fig. 4C and D). Utilizing the RF model, we identified the 15 most significant gene targets (Fig. 4E). Ultimately, a Venn diagram that consolidated the findings from the three methodologies pinpointed five definitive core targets linked to DEHP-induced colorectal cancer: CASP3, BCL6, BRD4, PPARA, and PRKCD (Fig. 4). Fig. 4Identification of essential target genes for DEHP-induced colorectal cancer (CRC) through machine learning techniques. (A, B) The Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression method was applied. The x-axis indicates the number of model genes associated with various (λ) values, with 15 genes recognized at the lowest λ value. (C, D) Essential genes were determined using the Support Vector Machine (SVM) coupled with Recursive Feature Elimination (RFE) method, resulting in the selection of 15 candidate genes. (E) Importance ranking of the Top 15 Genes as determined by the Random Forest (RF) model. (F) A Venn diagram illustrating the intersections among the Least Absolute Shrinkage and Selection Operator (LASSO), Support Vector Machine Recursive Feature Elimination (SVM-RFE), and Random Forest (RF) methods, highlighting shared core targets for DEHP-induced CRC Identification of essential target genes for DEHP-induced colorectal cancer (CRC) through machine learning techniques. (A, B) The Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression method was applied. The x-axis indicates the number of model genes associated with various (λ) values, with 15 genes recognized at the lowest λ value. (C, D) Essential genes were determined using the Support Vector Machine (SVM) coupled with Recursive Feature Elimination (RFE) method, resulting in the selection of 15 candidate genes. (E) Importance ranking of the Top 15 Genes as determined by the Random Forest (RF) model. (F) A Venn diagram illustrating the intersections among the Least Absolute Shrinkage and Selection Operator (LASSO), Support Vector Machine Recursive Feature Elimination (SVM-RFE), and Random Forest (RF) methods, highlighting shared core targets for DEHP-induced CRC Validating the expression of five key targets in psoriasis and diagnostic effectiveness examination We retrieved RNA-seq datasets pertaining to colorectal cancer patients (GSE21510 and GSE44076) from the GEO database to confirm the expression levels of key genes. In CRC tissues, the expression of CASP3, PPARA, and PRKCD was significantly lower compared to that in normal tissues, while BCL6 and BRD4 showed a notable increase (Fig. 5A). We generated ROC curves and computed the AUC to assess the diagnostic efficacy of these five core genes. The AUC values obtained for CASP3, BCL6, BRD4, PPARA, and PRKCD were 0.917, 0.823, 0.829, 0.809, and 0.845, respectively (Fig. 5B). These results highlight the significance of these five core targets in the development and prospective diagnosis of CRC. Fig. 5Analysis of RNA sequences concerning the expression of core genes in samples of colorectal cancer (CRC) compared to normal tissue. (A) Examination of the expression levels of five critical targets in biopsy samples obtained from CRC patients and healthy individuals (GSE21510 and GSE44076). (B) Evaluation of the diagnostic performance of these five core targets in differentiating between normal and CRC samples, conducted through receiver operating characteristic (ROC) curve analysis Analysis of RNA sequences concerning the expression of core genes in samples of colorectal cancer (CRC) compared to normal tissue. (A) Examination of the expression levels of five critical targets in biopsy samples obtained from CRC patients and healthy individuals (GSE21510 and GSE44076). (B) Evaluation of the diagnostic performance of these five core targets in differentiating between normal and CRC samples, conducted through receiver operating characteristic (ROC) curve analysis The correlation analysis between core genes and immune cells in CRC We applied the CIBERSORT algorithm to assess the distribution of 22 distinct immune cell types within both normal colorectal tissues and colorectal tumor tissues, while also examining their potential correlations with five critical genes: BCL6, BRD4, CASP3, PPARA, and PRKCD. Our analysis uncovered remarkable differences in the levels of infiltration from various immune cell types, including activated dendritic cells, M1 macrophages, neutrophils, activated CD4 + memory T cells, resting CD4 + memory T cells, gamma delta T cells, and activated mast cells (Fig. 6A and B). Additionally, we identified specific correlation patterns among these immune cell types (Fig. 6C). For example, activated CD4 + memory T cells showed a positive correlation with both M1 and M0 macrophages, indicating a potential cooperative role in the immune response. Meanwhile, gamma delta T cells were negatively correlated with naive B cells and dendritic cells, suggesting a complex interplay between these cell types. Notably, we found significant negative correlations between regulatory T cells (Tregs) and activated CD4 + memory T cells, as well as between M2 macrophages and activated mast cells, highlighting possible regulatory mechanisms in the tumor microenvironment. Following this, we examined the interactions between the five core genes and the infiltration of immune cells (Fig. 6D). Our results indicated significant positive correlations involving PPARA expression and the infiltration scores of Tregs, M2 macrophages, and neutrophils, which may suggest that PPARA plays a role in enhancing the presence of these immune subsets in the tumor environment. Additionally, we noted positive correlations between BRD4 and activated dendritic cells/M0 macrophages, along with associations between PRKCD and activated mast cells/neutrophils, further indicating the influence of these genes on immune cell behavior. Conversely, we observed negative correlations between CASP3 expression and the infiltration of activated CD8 + T cells and M1 macrophages, as well as an inverse relationship between BCL6 and populations of immunosuppressive cells, such as regulatory T cells. These findings indicate complex regulatory dynamics influenced by the expression of core genes within the immune landscape of colorectal tumors. Fig. 6Analysis of immune cell infiltration in colorectal cancer (CRC) versus healthy tissues utilizing the GSE21510 and GSE44076 datasets. (A–B) Application of the CIBERSORT algorithm to identify 21 distinct immune cell types present in both CRC and normal tissues. (C) Examination of the relationships among various immune cell types in CRC. (D) Investigation of the correlation between key target genes and immune cell infiltration levels in CRC Analysis of immune cell infiltration in colorectal cancer (CRC) versus healthy tissues utilizing the GSE21510 and GSE44076 datasets. (A–B) Application of the CIBERSORT algorithm to identify 21 distinct immune cell types present in both CRC and normal tissues. (C) Examination of the relationships among various immune cell types in CRC. (D) Investigation of the correlation between key target genes and immune cell infiltration levels in CRC Molecular Docking Verification of molecular docking was carried out between DEHP and the primary target proteins, utilizing the binding energy parameter as the baseline for determining effective docking. A greater absolute value of the binding energy parameter indicates a stronger binding affinity between the ligand and the target protein. The analysis of molecular docking focused on the interactions between DEHP and five key target proteins: BCL6, BRD4, CASP3, PPARA, and PRKCD. AutoDock and Pymol software facilitated the docking process, where a binding energy value lower than − 5 kcal/mol is typically interpreted as showing favorable binding activity. This research revealed that the minimum binding energies for DEHP with all six target proteins were below − 5 kcal/mol. More specifically, the minimum binding energy for DEHP with BCL6 was measured at -6.3 kcal/mol, -5.5 kcal/mol for BRD4, -6.4 kcal/mol for CASP3, -6.7 kcal/mol for PPARA, and − 5.3 kcal/mol for PRCKD (Fig. 7). The five docking outcomes demonstrate a robust attraction between the compounds and their respective targets, with PPARA exhibiting the lowest binding energy. Additionally, the binding energies for all proteins in association with DEHP remain below zero, implying that CRC can intrinsically interact with the core targets, thus highlighting their crucial involvement in the molecular mechanisms underlying DEHP-induced CRC. Fig. 7Molecular docking results of each hub target protein with DEHP. (a) DEHP -BCL6, (b) DEHP-CRD4, (c) DEHP-CASP3, (d)DEHP-PPARA and (e) DEHP-PRCKD Molecular docking results of each hub target protein with DEHP. (a) DEHP -BCL6, (b) DEHP-CRD4, (c) DEHP-CASP3, (d)DEHP-PPARA and (e) DEHP-PRCKD Molecular dynamics simulations and binding free energy assessment Although molecular docking analysis employed a semi-flexible docking approach, it cannot account for protein flexibility, structure, temperature, pressure, or solvent effects. Since DEHP-PPARA exhibited the lowest affinity in molecular docking, we further conducted MD simulations to verify its stability. RMSD analysis revealed that the complex system reached equilibrium after 50 ns, ultimately fluctuating around 2.1 Å, indicating high conformational stability (Fig. 8A). Rg results demonstrated overall structural compactness without significant expansion or contraction (Fig. 8B). SASA showed no notable changes after binding, further confirming the stable complex formation (Fig. 8C). Hydrogen bond analysis revealed an average of approximately 3 hydrogen bonds between the ligand and target protein, indicating favorable hydrogen bond interactions (Fig. 8D). The RMSF values were relatively low (mostly below 2.3 Å), suggesting low flexibility and high stability of the complex (Fig. 8E). The FEL displayed the free energy distribution based on RMSD and Rg, further illustrating the dynamic behavior of the system (Fig. 8F).In summary, this small molecule exhibits stable binding with the target protein and demonstrates favorable hydrogen bonding interactions. Fig. 8Molecular dynamics simulation of p DEHP-PPARA complex. A. RMSD values of DEHP-PPARA complexes over time; B. The Rg value of the DEHP-PPARA complex; C. SASA values of DEHP-PPARA complex; D. HBond values of DEHP-PPARA complex; E. RMSF value of the DEHP-PPARA complex. F; Free energy landscape map Molecular dynamics simulation of p DEHP-PPARA complex. A. RMSD values of DEHP-PPARA complexes over time; B. The Rg value of the DEHP-PPARA complex; C. SASA values of DEHP-PPARA complex; D. HBond values of DEHP-PPARA complex; E. RMSF value of the DEHP-PPARA complex. F; Free energy landscape map Discussion DEHP is a widely used plasticizer commonly found in food packaging, medical devices, and consumer goods [21]. Unfortunately, DEHP can leach from plastics and has become one of the prevalent environmental pollutants [22], posing risks to human health. It first garnered widespread attention due to its endocrine-disrupting properties, which may lead to a series of diseases affecting multiple organs, including the thyroid, testes, uterus, ovaries, liver, and nervous system [23]. In recent years, reports on the effects of long-term exposure to DEHP on the progression of various types of tumors have become increasingly common, with most studies focusing on liver cancer, breast cancer, and prostate cancer [24, 25]. However, little is known about whether DEHP affects the development of CRC, and its underlying molecular mechanisms remain unclear. To explore the causal relationship between plasticizers and CRC, we selected the widely used DEHP for this study. Predictions from two major toxicity prediction websites have indicated that DEHP exhibits significant carcinogenicity. Furthermore, research has also uncovered the carcinogenic potential associated with DEHP [26]. In this investigation, we discovered 57 potential toxicological targets linked to DEHP-induced CRC. Analyzing these targets through enrichment methods uncovered various pathways that might be affected by DEHP exposure, thereby emphasizing the molecular mechanisms implicated in CRC development. Initial GO and KEGG pathway assessments indicate that the occurrence of CRC could be influenced by DEHP exposure via several pathways, primarily those related to the cell cycle, apoptosis, and metastasis. These pathways can promote or suppress cancer, including the development of MicroRNAs in cancer, misregulation of transcription associated with cancer, and activation of receptors involved in chemical carcinogenesis. Research involving animals has demonstrated that multiple phthalates are capable of disrupting the immune system, for instance, by influencing the adjuvant characteristics of acquired immunity [27]. DEHP promotes the proliferation of T cells specific to antigens and subsequently increases the production of IL-4 and IFN-γ [28]. Earlier studies have shown that DEHP can impact the expression of markers related to differentiation and maturation on dendritic cells derived from peripheral blood mononuclear cells [29]. It is well-established that dendritic cells are the most effective cells in the human body for activating resting T cells, often referred to as professional antigen-presenting cells. Consequently, dendritic cells hold a crucial role in the body’s anti-tumor immune response [30]. Based on these insights, we propose that DEHP might significantly influence the onset and evolution of CRC by modulating dendritic cells and their associated signaling pathways. Through these mechanisms, DEHP affects cellular functions and stimulates tumor immune responses, leading to aggravated disease progression. We identified 15 potential core targets utilizing the LASSO algorithm, RF algorithm, and SVM-RFE model, and we found 5 core targets for DEHP-induced colorectal cancer by determining the intersection: CASP3, BCL6, BRD4, PPARA, and PRKCD. We assessed the levels of expression in the tissues of patients with CRC to explore more about the significance of these five primary targets in CRC. Our findings demonstrate that the expression of CASP3, PPARA, and PRKCD is notably downregulated in colorectal cancer tissues compared to normal tissues, while BCL6 and BRD4 show considerable upregulation. Moreover, we developed ROC curves to evaluate the potential diagnostic significance of these five core indicators for colorectal cancer. The analysis of the ROC curves indicated that the AUC for all five core targets exceeds 0.8 (CASP3: 0.917, BCL6: 0.823, BRD4: 0.829, PPARA: 0.809, and PRKCD: 0.845). These results imply that these five core targets have robust discriminatory capabilities and could act as valuable biomarkers for diagnosing CRC, thereby highlighting the essential role of the identified core targets in this research in the development of CRC. It is widely recognized that the immune system plays a critical role in the initiation and progression of solid tumors, including CRC [31]. In this study, we applied the CIBERSORT algorithm to RNA-seq data from normal and CRC tissues to estimate the infiltration of 21 immune cell types and to explore how the five DEHP-related hub genes correlate with the remodeled intestinal immune microenvironment. We observed that activated dendritic cells, M1 macrophages, neutrophils, activated and resting CD4⁺ memory T cells, γδ T cells, and activated mast cells were strongly associated with CRC status, in line with previous reports highlighting the importance of macrophages, neutrophils and CD4⁺ T-cell subsets in CRC progression and prognosis [32–34]. Within this altered microenvironment, the hub genes showed distinct and biologically plausible association patterns with immune cell subsets. PPARA expression was positively correlated with Tregs, M2 macrophages and neutrophils. This is consistent with the emerging view that PPAR-dependent lipid-metabolic programs favor the differentiation and maintenance of immunosuppressive T-cell and macrophage phenotypes, thereby dampening effective anti-tumor immune responses in the gut [35]. BRD4, a key epigenetic regulator in myeloid and T cells, was associated with activated dendritic cells and M0 macrophages [36].Given that BRD4 controls the transcription of inflammatory cytokines, costimulatory molecules and immune checkpoint ligands, its upregulation in dendritic and precursor macrophage populations may contribute to a chronically inflamed yet functionally tolerogenic tumor milieu that supports CRC growth and immune escape [37]. PRKCD, a ROS-sensitive kinase, correlated with activated mast cells and neutrophils; both cell types are major sources of reactive oxygen species, proteases and pro-tumorigenic mediators such as IL-6 and TNF-α, and PRKCD activation has been linked to NF-κB-driven inflammatory signaling and tissue remodeling, which may further promote angiogenesis and barrier disruption [38]. Conversely, CASP3 expression showed negative correlations with activated CD8⁺ T cells and M1 macrophages, suggesting that higher CASP3 activity may be associated with reduced infiltration or persistence of cytotoxic effector populations or with non-immunogenic forms of cell death that fail to efficiently recruit and activate these cells [39]. BCL6 exhibited inverse associations with certain immunosuppressive cell types, which is in line with its context-dependent role in shaping T-cell differentiation (e.g., follicular helper versus regulatory phenotypes) and germinal-center-like responses in the tumor microenvironment [40]. Taken together, these gene–immune patterns map onto known axes of apoptosis, epigenetic control and oxidative-stress-responsive signaling in CRC immunology and therefore support the biological plausibility of the correlations we observed. Together, these observations suggest that DEHP-induced dysregulation of the identified hub genes could contribute to re-wiring of the intestinal immune microenvironment, potentially shifting the balance towards a state characterized by increased Tregs and neutrophils and altered macrophage polarization, while limiting cytotoxic CD8⁺ T-cell and M1-macrophage activity. In conclusion, our study identified 57 putative toxicological targets that may link DEHP exposure to CRC, from which five hub genes (BCL6, BRD4, CASP3, PPARA, and PRKCD) were prioritized using an integrative machine-learning approach. Immune infiltration analysis revealed that these hub genes are significantly associated with multiple immune cell subsets, suggesting that DEHP may modulate the intestinal immune microenvironment through their dysregulation. Molecular docking demonstrated that DEHP can bind all five hubs, with the most favorable binding affinity observed for PPARA (− 6.7 kcal/mol), and this interaction was further supported by molecular dynamics simulations. ALL hub genes identified in our network are all well-established regulators of apoptosis, transcriptional control, and oxidative-stress–related signaling in CRC, supporting the overall biological relevance of our multi-target network. Among the five hubs, PPARA emerged as a particularly compelling candidate for in-depth mechanistic analysis. PPARA is a major transcriptional regulator of fatty acid oxidation and cellular energy metabolism, and DEHP has been reported to activate PPARA and induce mitochondrial dysfunction, endoplasmic reticulum stress and apoptosis in other tissues such as the testis [41]. In CRC, altered PPARA expression has been associated with lymph node metastasis and with a progressive downregulation from adjacent normal mucosa to advanced tumor stages [42], suggesting a role in tumor progression. In our datasets, PPARA showed consistent downregulation in CRC tissues and good diagnostic performance, and also displayed immunologically relevant correlations with Tregs, M2 macrophages and neutrophils. Building upon these observations, we selected PPARA as a representative hub gene for MD simulations following docking. The 100-ns explicit-solvent MD trajectory of the DEHP–PPARA complex revealed small and stable RMSD fluctuations after equilibration, largely constant radius of gyration and solvent-accessible surface area, and several persistent hydrogen bonds between DEHP and key residues in the PPARA ligand-binding domain. Free-energy landscape analysis further indicated a dominant low-energy basin, suggesting that the docked pose corresponds to a thermodynamically favorable binding mode rather than a docking artefact. These MD-based findings provide dynamic support for a direct DEHP–PPARA interaction under near-physiological conditions and substantially strengthen the mechanistic plausibility of a DEHP–PPARA axis in CRC beyond static docking scores alone. In our analysis of enriched pathways, we found that the KEGG pathway “Transcriptional misregulation in cancer” was significantly enriched. However, we concentrated on specific sub-pathways, including Polycomb group-associated repression, and nuclear receptor-associated transcriptional activation. These transcriptional regulatory mechanisms are well-established in CRC and correspond with the known molecular effects of DEHP. Furthermore, our enrichment analysis indicated that DEHP-induced carcinogenicity is associated with microRNAs (miRNAs). MiRNAs play a dual role as either oncogenes or tumor suppressors by influencing tumor-related signaling pathways, cell cycle regulation, apoptosis, metastasis, and drug resistance. The dysregulation of miRNA expression can lead to the suppression of tumor suppressor genes, promotion of genomic instability, induction of EMT, and metastasis, which are common causes of malignant tumors. MiRNAs regulate protein-coding gene expression by inhibiting translation or degrading target mRNA, thereby impacting the onset and progression of cancer [43]. Sun et al. reported that miRNA-1271 can inhibit CRC cell proliferation and invasion by negatively regulating metadherin&Wnt signaling [44]. Similarly, Zheng et al. indicated that miR-490-3p, acting as a suppressor, mediates CRC progression by regulating the Wnt/β-catenin signaling pathway [45]. These findings lead us to conclude that DEHP may induce CRC through multiple pathways. Taken together, our work underscores the value of integrating network toxicology, machine learning and bioinformatics with structural modelling to generate mechanistic hypotheses about how environmental pollutants may contribute to cancer. Compared with conventional animal experiments, which are time-consuming and costly, this in silico framework provides a scalable way to screen environmental chemicals, prioritize key targets and pathways, and highlight candidate mechanisms—such as the DEHP–PPARA axis and immune microenvironment remodeling—for focused experimental follow-up. By explicitly connecting DEHP, established CRC regulators (CASP3, BRD4, BCL6, PPARA, PRKCD), and immune cell infiltration patterns within a unified network, the present study adds an environmental toxicology perspective to existing knowledge on CRC biology and suggests that chronic DEHP exposure may reshape known oncogenic circuits rather than create entirely new ones. Although this study provides an integrated in silico framework linking DEHP-related toxicological targets to CRC-associated genes, several limitations must be acknowledged. First, all findings were generated entirely from computational analyses, including database-driven target mining, machine-learning–based feature selection, immune infiltration estimation, and molecular docking and MD simulations. As such, the results should be interpreted as hypothesis-generating rather than mechanistically conclusive. Second, the absence of experimental validation—such as DEHP exposure assays in CRC cell lines, functional perturbation of the core genes, or in vivo models evaluating tumor initiation and immune-microenvironment remodeling—limits the strength of causal inference. Third, although two independent GEO datasets were used for external validation, additional patient cohorts and prospective epidemiological data incorporating biomonitored DEHP levels would further enhance the generalizability of our findings. Future studies should therefore prioritize [1] validating the regulatory effects of DEHP on CASP3, BRD4, BCL6, PPARA, and PRKCD in biological systems [2], experimentally assessing whether the DEHP–PPARA interaction observed in docking and MD simulations translates into functional consequences in CRC cells, and [3] integrating laboratory, toxicological, and epidemiological evidence to build a multi-layer mechanistic understanding of how DEHP may contribute to CRC. Conclusion In summary, this study integrates network toxicology, machine-learning algorithms, transcriptomic validation, immune-infiltration profiling, and molecular docking combined with MD simulations to explore potential molecular links between DEHP exposure and CRC. Rather than proposing new oncogenic drivers, our findings re-position several CRC-associated genes as potential mediators through which DEHP may influence colorectal tumor biology. While these results offer a biologically plausible and environmentally focused framework for understanding the toxicological relevance of DEHP, they remain preliminary and require extensive experimental validation. Overall, this work provides a hypothesis-generating foundation for future mechanistic and translational studies on environmental pollutant exposure and CRC risk. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2