Computational Analysis Links Diethyl Phthalate to Potential Rheumatoid Arthritis Pathways

In a newly published computational investigation, researchers have deployed an integrated analytical framework combining network toxicology, interpretable machine learning, and structural molecular docking to identify candidate molecular networks connecting diethyl phthalate exposure to rheumatoid arthritis pathogenesis. Diethyl phthalate, a ubiquitous chemical compound widely utilized across everyday consumer products, has previously been associated with an elevated risk of developing rheumatoid arthritis, yet the underlying biological and molecular mechanisms driving this observed statistical link have remained unclear. To explore these putative computational links, the investigation systematically delineated chemical-related molecular networks and screened critical rheumatoid arthritis targets that could potentially mediate the compound’s influence on disease processes. The resulting computational findings highlight prospective interactions involving immune inflammation, metabolic regulation, and programmed cell death, offering initial hypothesis-generating evidence regarding how common environmental chemical exposures might interface with chronic autoimmune vulnerability.

The investigative workflow was constructed upon a systematic network toxicology framework designed to screen and delineate candidate disease targets by integrating chemical target prediction directly with transcriptomic data mining. The researchers analyzed rheumatoid arthritis transcriptomic datasets, applying formal batch correction procedures across the expression data to mitigate technical variation and ensure cross-cohort consistency. Following the completion of batch correction, the team performed differential expression analysis in conjunction with weighted gene co-expression network analysis, widely designated as WGCNA, to characterize coordinated patterns of gene expression within the disease cohorts. By mapping the intersection between computationally predicted diethyl phthalate targets and the transcriptomic signatures extracted from rheumatoid arthritis samples, the computational pipeline successfully pinpointed an overlapping set of sixty-nine shared targets for further downstream functional and algorithmic prioritization.

Following the identification of the sixty-nine overlapping targets, the researchers conducted detailed functional enrichment analyses on these intersecting molecular elements to determine which biological processes were predominantly implicated. The resulting functional enrichment profiles indicated that these intersecting targets are strongly linked to critical pathological mechanisms, most notably immune inflammation, biological pathways governing programmed cell death, and cellular metabolic regulation. By systematically mapping these overlapping molecular components to defined cellular functions, the analytical process established a computational basis for understanding how chemical perturbation by diethyl phthalate could potentially interact with inflammatory signaling cascades and cellular survival mechanisms that characterize the development and progression of rheumatoid arthritis.

To prioritize the most critical core targets from the sixty-nine candidate genes, the research team constructed and evaluated multiple machine-learning models. Among the evaluated computational architectures, a sequential machine-learning framework combining least absolute shrinkage and selection operator regression with random forest algorithms, termed the LASSO-RF sequential framework, achieved optimal predictive discrimination. In the training dataset, this sequential modeling strategy demonstrated exceptional discriminatory capability, attaining an apparent area under the receiver operating characteristic curve of 0.9998, with a ninety-five percent confidence interval spanning from 0.999 to 1.000. To assess the generalizability and robustness of this model beyond the primary training cohort, the framework underwent cross-cohort validation across independent external transcriptomic datasets. In the GSE55235 validation cohort, the model sustained robust diagnostic discrimination, yielding an area under the curve of 0.830 with a ninety-five percent confidence interval between 0.647 and 1.000. Similarly, evaluation within the independent GSE77298 cohort yielded an area under the curve of 0.839 with a ninety-five percent confidence interval ranging from 0.655 to 1.000.

To address the inherent complexity of advanced predictive algorithms and ensure transparent model interpretation, the researchers implemented SHapley Additive exPlanations, commonly known as SHAP, to determine the individual predictive contribution of each feature to the model’s classifications. Through this interpretable machine-learning approach, the investigators successfully prioritized six primary hub genes: NLRP1, ADORA2A, XBP1, PPP1CA, SAE1, and PIN1. The functional enrichment analyses associated these specific hub genes directly with inflammatory cascades, programmed cell death regulation, and metabolic pathways, underscoring their prospective involvement in the molecular alterations that occur when cellular systems encounter diethyl phthalate in the context of autoimmune susceptibility.

To computationally evaluate the physical feasibility of direct chemical interaction between diethyl phthalate and the prioritized hub targets, the researchers conducted molecular docking simulations alongside molecular dynamics analyses. The molecular docking assessments indicated thermodynamically plausible binding interactions between the diethyl phthalate ligand and four of the identified core targets: ADORA2A, NLRP1, PIN1, and PPP1CA. To further probe the physical behavior and structural stability of these predicted associations under simulated physiological conditions, the investigators executed a 100-nanosecond molecular dynamics simulation focusing specifically on the diethyl phthalate and PIN1 complex. This extended molecular dynamics trajectory provided computational support for the conformational stability of the DEP–PIN1 complex, demonstrating that the ligand-protein assembly maintained structural coherence across the simulated temporal window.

In their concluding assessment, the authors emphasized that their integrative computational architecture underscores putative mechanistic links connecting diethyl phthalate exposure to critical biological pathways governing adenosine signaling and inflammasome activation. However, the researchers explicitly underscored several essential methodological limitations that frame the proper interpretation of these findings. Most significantly, the investigation was computational in design and was limited by the lack of empirical exposure data as well as the absence of experimental biological validation. Consequently, the investigators emphasized that these findings represent hypothesis-generating insights rather than definitive proof of a causal clinical relationship. The authors concluded that while experimental validation remains necessary, this computational framework establishes a valuable foundation for subsequent research aimed at elucidating the mechanistic interplay between environmental factors and rheumatoid arthritis risk.

Subject of Research: Computational investigation of molecular mechanisms connecting diethyl phthalate exposure to rheumatoid arthritis

Article Title: Integrating network toxicology, interpretable machine learning, and molecular docking to unravel key molecular networks in diethyl phthalate-related rheumatoid arthritis

Article References: Jiang, Q., Wu, X., Zhang, W., Li, M., Shao, L., Xie, J., Yang, J., Chen, T., Wang, B., Chen, Y., & Xue, L. (2026). Integrating network toxicology, interpretable machine learning, and molecular docking to unravel key molecular networks in diethyl phthalate-related rheumatoid arthritis. Arthritis Research & Therapy. https://doi.org/10.1186/s13075-026-03946-z

Image Credits: AI Generated

DOI: 10.1186/s13075-026-03946-z

Keywords: Diethyl phthalate, Rheumatoid arthritis, Network toxicology, Machine learning, Molecular docking, Molecular dynamics simulation, NLRP1, ADORA2A, XBP1, PPP1CA, SAE1, PIN1, SHAP, WGCNA, LASSO-RF, Adenosine signaling, Inflammasome activation, GSE55235, GSE77298