Journal: Frontiers in Pharmacology

Article Title: Integrative machine learning and network toxicology framework for assessing environmental pollutant TCDD-induced osteoarthritis risk: a computational cheminformatics approach

doi: 10.3389/fphar.2026.1811221

Figure Lengend Snippet: Machine learning-based identification and validation of hub genes. (A) AUC values were displayed across the training dataset and two validation cohorts ( GSE169077 and GSE178557 ). The rightmost column shows the mean AUC. The top-performing algorithms are highlighted. (B) ROC curves validating the diagnostic performance of the identified hub genes (PTPRC, PTGS1, CBR1, HTR2B, and PTGS2) in distinguishing OA samples from controls. (C) Volcano plot highlighting the expression patterns of the five hub genes among all differentially expressed genes. Red dots indicate upregulated genes, green dots indicate downregulated genes, and hub genes are labeled. (D) Bar plot showing the mean absolute SHAP (SHapley Additive exPlanations) values for each feature, indicating the relative importance of genes in the prediction model. (E) Correlation matrix showing pairwise Spearman correlations among the hub genes. Each scatter plot displays the relationship between two genes, with correlation coefficients shown (F) SHAP summary plot displaying the distribution of SHAP values for each hub gene across samples from two validation cohorts ( GSE169077 and GSE178557 ). Color indicates feature value (red: high, blue: low). (G) SHAP waterfall plot illustrating the contribution of individual features to a single prediction, demonstrating how each gene contributes to the final prediction outcome.

Article Snippet: Sections were incubated overnight at 4 °C with rabbit primary antibodies against PTGS1 (1:200, Cat# GB112399 , Servicebio, Wuhan, China) and PTGS2 (1:100, Cat# GB155672 , Servicebio, Wuhan, China), respectively.

Techniques: Biomarker Discovery, Diagnostic Assay, Expressing, Labeling

Journal: Frontiers in Pharmacology

Article Title: Integrative machine learning and network toxicology framework for assessing environmental pollutant TCDD-induced osteoarthritis risk: a computational cheminformatics approach

doi: 10.3389/fphar.2026.1811221

Figure Lengend Snippet: Comprehensive immune infiltration analysis of hub genes in OA. (A) Correlation network among immune cell types in OA. Heatmap depicts Spearman correlations among 22 immune cell types estimated by CIBERSORT in OA samples. Red indicates positive correlation; blue indicates negative correlation. Asterisks denote statistical significance (*p < 0.05, **p < 0.01, ***p < 0.001). (B) Hub gene–immune cell correlation analysis. Heatmap shows Spearman correlation coefficients between the expression levels of four hub genes (PTGS1, CBR1, HTR2B, PTGS2) and the relative proportions of 22 immune cell types. Red indicates positive correlation, blue indicates negative correlation. Color intensity corresponds to the absolute correlation coefficient. (C) Significant hub gene–immune cell correlations. Lollipop plot displays all significant gene–immune cell pairs (|r| > 0.3, FDR <0.05). The x-axis represents Spearman correlation coefficient; the y-axis shows gene–immune cell pairs sorted by correlation strength. Dot size indicates −log10(FDR). (D) Representative scatter plots of significant gene–immune cell pairs. Six significant correlations are displayed: PTGS2 vs. Mast cells activated (r = 0.726, FDR <0.001), PTGS2 vs. Mast cells resting (r = −0.604, FDR <0.001), PTGS2 vs. T cells CD4 memory resting (r = 0.531, FDR <0.001), CBR1 vs. T cells CD4 memory resting (r = −0.491, FDR <0.001), CBR1 vs. Mast cells resting (r = 0.442, FDR <0.001), and HTR2B vs. Mast cells activated (r = −0.408, FDR <0.001). (E) Differential immune cell infiltration between control and OA groups. Box plots compare the estimated proportions of 22 immune cell types between control (n = 37) and OA (n = 34) samples. Statistical significance was assessed using Wilcoxon rank-sum test (*p < 0.05, **p < 0.01, ***p < 0.001).

Article Snippet: Sections were incubated overnight at 4 °C with rabbit primary antibodies against PTGS1 (1:200, Cat# GB112399 , Servicebio, Wuhan, China) and PTGS2 (1:100, Cat# GB155672 , Servicebio, Wuhan, China), respectively.

Techniques: Expressing, Control

Journal: Frontiers in Pharmacology

Article Title: Integrative machine learning and network toxicology framework for assessing environmental pollutant TCDD-induced osteoarthritis risk: a computational cheminformatics approach

doi: 10.3389/fphar.2026.1811221

Figure Lengend Snippet: Molecular dynamics simulation analysis of TCDD binding to four hub proteins. (A–D) Representative three-dimensional structures of four hub proteins in complex with TCDD after MD simulations: (A) HTR2B, (B) CBR1, (C) PTGS1, and (D) PTGS2. Proteins are shown in cartoon representation colored by secondary structure, and TCDD is displayed as stick representation. (A2–D2) RMSF profiles of protein residues for each protein–TCDD complex during MD simulations. RMSF values reflect residue-level flexibility, with higher fluctuations mainly observed in terminal regions and solvent-exposed loops (A3–D3) FELs of the four protein–TCDD complexes projected onto the first two principal components (PC1 and PC2) obtained from principal component analysis (PCA) of the MD trajectories. Color gradients indicate free energy levels (kcal/mol), where blue regions represent low-energy, stable conformational states and red regions represent high-energy states. (E) RMSD of protein backbones for all four protein–TCDD complexes over the simulation time, illustrating the overall structural stability of each protein upon ligand binding. (F) RMSD of TCDD relative to the protein binding site for all four complexes, reflecting the positional stability of the ligand during MD simulations. (G) RT-qPCR validation of PTGS1 and PTGS2 mRNA expression levels in OA samples and healthy controls (n = 10 per group) (H) Representative IHC staining images of PTGS1 and PTGS2 protein expression in synovial tissue sections from OA patients and healthy controls. (*p < 0.05, **p < 0.01, ***p < 0.001).

Article Snippet: Sections were incubated overnight at 4 °C with rabbit primary antibodies against PTGS1 (1:200, Cat# GB112399 , Servicebio, Wuhan, China) and PTGS2 (1:100, Cat# GB155672 , Servicebio, Wuhan, China), respectively.

Techniques: Binding Assay, Residue, Solvent, Ligand Binding Assay, Protein Binding, Quantitative RT-PCR, Biomarker Discovery, Expressing, Immunohistochemistry