Prognostic Model Of Serous Ovarian Cancer Based On Gene Relative Expression Orderings

Objective:Ovarian cancer has the highest mortality rate among gynecological tumors.To improve the survival rate of ovarian cancer,researchers are committed to exploring individualized prognostic markers with clinical applications.Currently,most prognostic genetic marker models calculate risk scores based on the aggregation of gene expression levels.However,due to influences such as batch effects and platform differences,scoring-based models are prone to produce spurious risk classifications in independent samples,making them difficult to apply.This study aimed to construct the best prognostic model for overall survival in serous ovarian cancer based on the relative order relationships between genes or pathways within the samples.Methods:In this study,eight serous ovarian cancer prognostic datasets were collected,three of which were combined into the training set,including two GEO datasets(GSE18520,GSE19829)and one TCGA dataset.One set is the test set(GSE13876);The remaining four sets are validation sets(GSE14764,GSE26712,GSE26193,GSE53963).Firstly,the Cox regression model is used to identify prognosis-related genes from the training set,and a gene pair matrix is constructed according to the relative order relationship of their expression values.Then,the greedy algorithm is used to search for the best predictive gene pair combination,and then the test set identifies the group with the largest C-index value in the gene pair combination as the final prognostic marker gene pair.Finally,the classification efficiency of prognostic gene markers in the training set(561 cases)and the three validation sets(185,78 and 174 cases)was evaluated by the ROC curve.The pathway enrichment score of each sample was obtained by using the gene set enrichment analysis of a single sample,and the pathway pair matrix was constructed according to the relative order relationship of the score value.Then,a pathway prediction model was constructed by a method similar to the above gene pair prognostic model and its predictive efficacy was tested.For the final gene pair prediction model,gene enrichment,differential gene pathway enrichment analysis between high and low-risk groups,differential analysis of immune cell infiltration between two groups,analysis of different immunotherapy responses of high and low-risk groups,analysis of response to chemotherapy drugs in high and low-risk groups and development based on prognostic gene marker nomogram were carried out.Results:This study screened a prognostic biomarker(SOV-GP20)consisting of 20 gene pairs.Kaplan-Meier analysis showed that the mean area(AUC)values under the curve for SOV-GP20 prediction of serous ovarian cancer were 0.756,0.590,0.630,and 0.680,respectively,in the training dataset and independent validation datasets from different platforms.There were statistically significant differences in the overall survival of the high-and low-risk groups for serous ovarian cancer(p<0.05,Wilcoxon test).The forest plot shows that SOV-GP20 is an independent prognostic factor for serous ovarian cancer(p<0.05,log-rank test).For the pathway-based screening,a prognostic biomarker(SOV-PP12)consisting of 12 pathway pairs was obtained.The KaplanMeier analysis showed that there was a statistically significant difference in overall survival between the high-and low-risk groups stratified by SOV-PP12 in the training dataset(p<0.05,Wilcoxon test)and no significant difference in overall survival in the other three validation sets.Genes in SOV-GP20 are significantly enriched in biological pathways related to regulating cytokine production,viral entry into host cells,and tumor recurrence(p<0.05,Wilcoxon test).In the training set,the differentially expressed genes predicted by SOV-GP20 between samples in the high-and low-risk groups were significantly enriched in the PI3K-Akt signaling pathway,proteoglycan pathway,and AGE-RAGE signaling pathway.Also,there were significant differences in the distribution of CD8+T cells,neutrophils,macrophages,and dendritic cells between the high-and low-risk groups(P<0.001,Wilcoxon test).The results of immunotherapy response analysis showed that common immune test nodes(CTLA4,PDCD1,LAG3,TNFRSF14,CD48,MICB)had significant differential expression between high-and low-risk groups(P<0.05,Wilcoxon test).Drug susceptibility analysis showed significant differences between chemotherapeutic responses in the high-and low-risk groups predicted by SOV-GP20 in the TCGA dataset(P<0.05,Wilcoxon test).Moreover,throughout the training set,cisplatin,gemcitabine,oxaliplatin,sorafenib,tamoxifen,and topotecan were more effective in the low-risk group.In contrast,5-fluorouracil was more effective in patients in the high-risk group,with significant differences between highand low-risk groups(P<0.05,Wilcoxon test).The clinical fitness nomogram showed that the SOV-GP20 risk score was more accurate than other clinical factors.In addition,SOV-GP20 had the highest mean C-index value(0.624)compared to seven previously established predictive prognostic models for serous ovarian cancer.Conclusion:Compared with the traditional method of risk scoring based on expression level aggregation,the method based on the relative order relationship within the sample is more robust.The prognostic marker gene model screened by this method is expected to be applied to the prognosis of individualized tumors of serous ovarian cancer and has significant clinical value for the individualized treatment of serous ovarian cancer.