Integrating bulk RNA-seq and scRNA-seq analyses with machine learning to predict platinum response and prognosis in ovarian cancer.

Platinum-based therapy is an integral part of the standard treatment for ovarian cancer. However, despite extensive research spanning several decades, the identification of dependable predictive biomarkers for platinum response in clinical practice has proven to be a formidable challenge. Recently, the development of single-cell technology has enabled more precise investigations into the heterogeneity of cancer. In this study, we isolated cancer cells from the single-cell transcriptomic data of platinum-sensitive and platinum-resistant patients with ovarian cancer. Differential gene analysis of platinum-sensitive and platinum-resistant cancer cells revealed that several of the differentially expressed genes had previously been reported in other studies to be associated with platinum resistant. Gene set enrichment analysis revealed the up-regulation of pathways involved in processes such as autophagy, cell cycle regulation, and DNA damage repair, which are known to promote platinum resistance in ovarian cancer. Based on these findings, we hypothesized that these differentially expressed genes could be used to predict the response of ovarian cancer patients to platinum-based chemotherapy. To validate this hypothesis, we explored 7 different machine learning models for predicting platinum chemotherapy response at varying feature gene counts. Ultimately, the random forest model performed the best, with 5 genes (PAX2, TFPI2, APOA1, ADIRF and CRISP3) and achieve an AUC of 0.993 in test cohort and 0.989 in GSE63885 independent validation cohorts. We named this model GPPS (Genes to Predict Platinum response Signature). Furthermore, we discovered that the GPPS model can also predict patient prognosis.