Abstract
Nicotinamide metabolism play important roles in the formation and progression of ovarian cancer(OC). This study aimed to develop a prognostic marker related to nicotinamide metabolism in OC based on multiple machine learning. Gene expression profiles were obtained from the TCGA and GEO database. The TCGA dataset was used as a train cohort and GSE19829 was used as a validation cohort. The levels of these feature genes were also confirmed in an independent single-cell dataset GSE147082. A consensus prognostic model was constructed from a combination of 112 machine learning algorithms, including 10 classical ones. The STRING database facilitated the construction of protein-protein interaction networks around hub genes. Survival analysis was performed using the Kaplan-Meier method with the survival R package. Gene Set Enrichment Analysis (GSEA) was conducted via the cluster Profiler R package, while mutation patterns were examined with maftools. Immune infiltration and regulatory differences were assessed using the estimate R package. Single-cell datasets underwent quality control and cell-type annotation with Seurat v4.0, and pseudotime trajectory analyses were executed to map cellular development. Our results identified stable prognostic genes associated with OC and developed a risk scoring system that effectively stratifies patients into high- and low-risk groups with significant survival differences (p<0.05). A final prognostic model based on 23 hub genes demonstrated a robust mean concordance index (C-index) of 0.709. Furthermore, we revealed significant disparities in immune checkpoint markers expression between different risk groups, underscoring potential therapeutic implications. Finally, the model was tested on additional cancer types and clinical utility was assessed using calibration and decision curve analysis. Overall, this novel multiple machine learning could develop a prognostic marker, related to Nicotinamide metabolism, which is very promising in clinical promotion.