-
Expression of the 23 m6A-related genes in the differential genes was extracted.
We used the limma R package to select a total of 23 m6A-related genes in the differentially expressed gene dataset(Figure.1a), and we screened out 11 m6A-related genes with expression differences(Figure.1b), including R BM15B, WTAP, YTHDF2, METTL14, RBM15, YTHDF3, IGF2BP1, YTHDC2, LRPPRC, METTL3 and HNRNPA2B1. We found that WTAP, M ETTL14, R BM15, R BM15B, YTHDF2, and YTHDF3 were highly expressed in the experimental group (CD), and the rest were highly expressed in the healthy ileal control group.
2. Construction Of The Rf Model And The Svm Model
RF and SVM models were established to select candidate M6A-related genes from 23 M6A-related genes to predict the occurrence of Crohn's disease. We found the random forest tree models with smaller residuals by the "reverse cumulative distribution of residuals" and the "residue boxplot"(Figure.1c-d). Therefore, we considered the RF model the best model to predict the occurrence of Crohn's disease. After ranking these genes according to importance, we visualized the 11 m6A-related genes.
As shown in Figure.1f, the abscissa for the number of trees, the ordinate is the cross-validation error. Further, the red line represents the error of the experimental group (Crohn's disease group), the green line represents the error of the control group, and the black line represents the error of all samples(Figure.1h). We found a slight error of 106 for the optimal number of trees in the curve and then reconstructed the forest tree model. Genes are of more importance and significance than two were selected as signature genes(Figure.1g). We found that the number of m6A regulators in the top five positions (FMR1, KIAA1429, WTAP, YTHDC2, and ZC3H13) were selected as candidate genes. We found that the number of m6A regulators in the top five (FMR1, KIAA1429, WTAP, YTHDC2 and ZC3H13) were selected as candidate genes. Finally, the ROC curve was drawn to evaluate the model(Figure.1e), and its AUC value also shows that the RF model has higher accuracy than the SVM model.
3. Construction Of Nomogram Model
The "RMS" R package used constructed nomogram models based on five candidate M6A-related genes to predict the prevalence of Crohn's disease(Figure.1k). The calibration curve shows that the nomogram model is accurate in its prediction(Figure.1i). In the DCA curve(Figure.1j), we found that the red lines mainly remained above the gray and black lines on a scale from 0 to 1, indicating that decision-making based on nomogram models may benefit patients with Crohn's disease. On the clinical impact curve, we found that the nomogram model's predictive power was very significant.
4. Differences In Connections Between Different Molecular Subtypes
We divided the samples into cluster A and B subtypes based on the 11 most critical m6A-related genes (Figure.2a). Among them, subtype A included 129 cases, and subtype B included 214 cases. We subsequently showed the difference in the expression situation of the essential m6A-related genes between the two isoforms by heatmaps and histograms, and we found that the expression was higher in isoform A than in isoform B. Conversely, RBM15 and YTDHF3 were not significantly different in isoforms A and B. We found in PCA that nine essential m6A-related genes could completely distinguish between the two m6A patterns(Figure.2b). A total of 315 m6A were selected between the two m6A patterns, with associated differentially expressed genes. We applied to GO functional enrichment analysis to obtain possible mechanisms for these differential genes in Crohn's disease and to visualize the results with histograms(Figure.2f-i). We found that GO: 0015711, GO: 0015849, and GO: 0046942 correlated with ion transport.
When ssGSEA calculated immune cell abundance in Crohn's disease samples and evaluated the correlation between 11 important m6A modulators and immune cells(Figure.2c-d), we found that HNRNPA2B1 was positively associated with many immune cells. We explored the differential immune cell infiltration between patients with high and low HNRNPA2B1 expression(Figure.2e). The results showed that patients with Crohn's disease with high HNRNPA2B1 expression were immune immunity.
Finally, we analyzed the differential immune cell infiltration between the two m6A patterns. We found that cluster A was associated with MDSC immunity, while cluster B was associated with monocyte immunity.
5.identification Of Two Different M6a Gene Patterns And The Generation Of M6a Signature Genes
Based on 315 M6A-related differentially expressed genes(Figure.3a), Crohn's disease patients were divided into different genomic subtypes using consensus clustering. Consistent with grouping M6A-related gene subtypes, we found two distinct genotypes of m6A (gene cluster A and gene cluster B). Figure.3b shows the expression levels of 11 differentially expressed genes related to m6A in gene clusters A and B.As shown in Figure.3c, the differential expression levels of 11 crucial m6A regulators and immune cell infiltration between gene cluster A and gene cluster B are also contrary to the results of M6A-related gene grouping subtypes, which again validates the accuracy of our grouping by consensus clustering method. To quantify the m6A pattern, we used the PCA algorithm to calculate the m6A score for each sample. Then the m6A scores of m6A pattern typing and M6A-related differential gene pattern typing were compared. It was shown that a higher m6A score was found in subtype B than in subtype A in m6A pattern typing, while in M6A-related genotyping, a higher m6A score was found in subtype A than in subtype B. The relationship between m6A methylation-related gene pattern typing, m6A gene pattern and m6A score was visualized in the Sankey plot. It was found that there was a negative regulatory relationship between different genotypes(Figure.3d-e).
6. The Role Of The M6a Molecular Subtype In Differentiating Crohn'S Disease
To further reveal the relationship between the m6A grouping subtypes and Crohn, we investigated the correlation between the m6A patterns and gene cluster(Figure.4a). We found that PHOX2B and ATG16L were higher than genotyping B, while NCF4, I L-33, and NOD2 were higher than genotyping B than genotyping A. In m6A typing A, NCF4, I L-33, and NOD2 were higher than B, while PHOX2B and ATG16L were higher than genotyping A. PHOX2B, NCF4, and NOD-2 have been closely linked to the development of Crohn's disease(Figure.4b-c).