Network toxicology and machine learning reveal the toxicological impact of Bisphenol A exposure on osteoarthritis

Osteoarthritis (OA) is the most common chronic degenerative joint disease, and increasing evidence suggests that environmental exposures play a crucial role in its onset and progression. Bisphenol A (BPA), a widespread environmental endocrine disruptor, has been associated with inflammation, oxidative stress, and abnormal bone metabolism; however, its potential mechanistic involvement in OA development remains unclear. In this study, BPA targets were 1st predicted using the CHEMBL, SwissTargetPrediction, and Similarity Ensemble Approach databases. Eight OA transcriptomic datasets were retrieved and integrated from the Gene Expression Omnibus database to construct training and validation cohorts. Differential expression analysis and weighted gene co-expression network analysis were performed on the training cohort, and their union was used to identify OA-related key genes. BPA targets were intersected with OA key genes to establish a BPA-OA interaction network using Cytoscape, followed by functional enrichment analyses. Based on the intersecting genes, 113 machine learning models were applied to identify the optimal predictive model, and SHapley Additive exPlanations analysis was conducted to interpret feature contributions. Core genes were subsequently identified and validated through molecular docking to assess their binding stability with BPA. A total of 235 potential BPA targets were predicted across the 3 databases. From the Gene Expression Omnibus database, 8 datasets were obtained, of which 5 were integrated as the training set and 3 as the validation set. In the training set, 541 differentially expressed genes were identified, and weighted gene co-expression network analysis yielded 3575 genes; their union resulted in 3739 OA-related genes. The intersection of these genes with BPA targets produced 47 candidate genes. Functional enrichment analysis revealed significant involvement in cytokine signaling, calcium signaling, lipid metabolism, and TRP channel-related pathways. Among the 113 machine learning models, the plsRglm ensemble model performed best (mean area under the receiver operating characteristic curve = 0.901). SHapley Additive exPlanations analysis further identified CLK1, PTPRC, and ALDH5A1 as core targets. Molecular docking confirmed stable binding of BPA to these proteins, with binding energies below -5 kcal/mol. This study systematically elucidates, for the 1st time, the potential mechanisms by which BPA may contribute to OA progression via inflammation- and oxidative stress-related pathways, using a network toxicology and machine learning framework. Furthermore, CLK1, PTPRC, and ALDH5A1 were identified as key targets, providing novel insights into the environmental toxicology of OA and potential therapeutic targets.