Fig 2: Establishment of a diagnostic model of NAFLD. (A) Venn diagram shows the coexisting genes (ABCC5, DHODH, TUBG1) of the diagnosis-related gene module darkturquoise in WGCNA and common differential genes. (B) Visualizing the expression profile of ABCC5 and TUBG1 in NAFLD by a heat map. (C) LASSO model tuning parameters were selected through 10-fold cross-validation. (D) LASSO coefficient spectra of diagnosis-relevant gene sets with dotted lines representing values selected by 10 cross-validations. A diagnostic model containing two genes (ABCC5 and TUBG1) was obtained. (E) Box plot presenting the difference in the risk score between the NAFLD group and the normal group. (F) ROC curve confirms that the diagnostic model has high diagnostic efficiency (AUC = 0.893). (G) Based on the new dataset (GSE37031), the risk score was calculated according to the model; the risk value in the NAFLD group was significantly higher than that in the normal group. (H) The expression level of TUBG1 in the GSE37031 dataset visualized by box plot in the NAFLD group and normal liver tissue sample group (P<0.05 is considered statistically significant). (I) Box plot visualization of ABCC5 expression in the GSE37031 dataset between the NAFLD group and the normal group.

Fig 3: Construction of hepatocellular carcinoma prognostic model. (A) Univariate Cox analysis suggested that ABCC5 and TUBG1 were correlated with the prognosis of HCC. Multivariate Cox analysis suggested that ABCC5 and TUBG1 were independent risk factors (red represents HR>1; dark blue represents HR<1, P<0.05 is statistically significant). (B) Both ABCC5 and TUBG were upregulated in the high-risk group (red represents high-risk group; dark blue represents low-risk group). (C) The risk value was calculated by multivariate Cox regression. With the increase of the risk value, the survival time decreased, and the number of patients who died increased. (D) With risk score as the observed value, the time ROC curve confirmed that it had good diagnostic value at 1-year, 2-year, and 3-year survival nodes (the AUC of a 1-year ROC curve was 0.8, the 2-year ROC curve was 0.68, and the 3-year ROC curve was 0.68). (E) The box plot shows that the death group has higher risk scores (0 represents alive group; 1 represents dead group).

Fig 4: Immune infiltration and ceRNA. (A) Heat map showing pathways that differ significantly between the normal and NAFLD groups (red for NAFLD group and dark blue for normal group). (B) Heat map shows pathways with significant differences between the normal group and the HCC group (red represents the cancer group; dark blue represents the normal group). (C) Venn diagram shows coupregulated and codownregulated pathways in both diseases. (D) The upper box plot shows the immune infiltration of 28 cells in the high- and low-risk group of NAFLD, and the lower box diagram shows the immune infiltration of 28 cells in in the high- and low-risk group of HCC. (E) ABCC5 and TUBG1-related ceRNA networks were constructed. Red represents mRNA, purple represents microRNA, and yellow represents lncRNA.

Fig 5: Risk model was tested with the ICGC-HCC dataset. (A) The box plot shows that the expression of ABCC5 and TUBG in the ICGC-HCC dataset weas significantly higher in the HCC group. (B) The box plot shows that in the ICGC-HCC dataset, the risk score in the HCC group was significantly higher than that in the normal group. (C) With the risk score as the observed value, the timeROC curve confirmed that our prognostic model still had a high diagnostic value (the AUC of the 1-year ROC curve was 0.7, the 2-year ROC curve was 0.66, and the 3-year ROC curve was 0.74). (D) The survival curve suggested that patients in the high-risk group had a poor prognosis (red represents high-risk group; yellow represents low-risk group, P<0.05 is statistically significant).