Identification of platinum-resistance related small GTPase binding signatures to predict the prognosis of ovarian cancer by machine learning and integrated bioinformatic analyses

Abstract

Background High grade serous ovarian cancer (HGSOC) had high lethality due to its high relapse rate and acquired drug resistance. The tumor immune microenvironment (TIME) of HGSOC was heterogeneous, and mostly immune cold. We aimed to build the bridge between platinum-resistance related signatures and patient overall survival (OS). Methods The RNA sequencing data from GSE160626 was used for extraction of platinum-resistance related genes. The TCGA-OV cohort were fitted into 101 kinds of machine learning methods, and the validation cohort included GSE9899, GSE63885 and GSE26193. Numerous methods including the Cindex, receiver operating characteristic curve (ROC), univariate and multivariate Cox regression, and the decision curve analysis (DCA) were applied to detect the performances of platinum-resistance related risk score (PRRS) and a PRRS based nomogram. The single-cell RNA sequencing data and Spatial Transcriptomics data were used to determine the risky cell types correlated with our PRRS. Results Based on platinum-resistance related genes, we conducted consensus clustering and defined a platinum-resistance resembling cluster, which had significantly shorter OS. And with DEGs related to small GTPase between two clusters, we established a PRRS and a PRRS based nomogram, which had excellent performances in predicting OS of serous ovarian cancer patients. We further determined SPP1 + M2-like Macrophages were risky factors correlated with the PRRS, and determined ABCA1 and NDRG1 as the hub genes related to patient OS. Conclusion Small GTPase was a dominant feature of platinum-resistance resembling clusters. PRRS had terrific predicting value and correlated with SPP1 + M2-like Macrophages.