Multiomics evaluation and machine learning optimize molecular classification, prediction of prognosis and immunotherapy response for ovarian cancer

Abstract

Background

Ovarian cancer (OC), owing to its substantial heterogeneity and high invasiveness, has historically been devoid of precise, individualized treatment options. This study aimed to establish integrated consensus subtypes of OC using different multiomics integration methodologies.

Methods

We integrated five distinct multiomics datasets from multicentric cohorts to identify high-resolution molecular subgroups using a combination of 10 and 101 clustering and machine learning algorithms, respectively, to develop a robust consensus multiomics-related machine learning signature (CMMS).

Results

Two cancer subtypes with prognostic significance were identified using multiomics clustering analysis. 10 essential genes were identified in the CMMS. Patients in the high CMMS group exhibited a poorer prognosis, with a “cold tumor” phenotype and an immunosuppressive state with reduced immune cell infiltration. In contrast, patients in the low CMMS group exhibited a more favorable prognosis, with immune activation and a “hot tumor" phenotype characterized by increased tumor mutation burden, tumor neoantigen burden, PD-L1 expression, and enriched M1 macrophages. Eight independent immunotherapy datasets were validated to further corroborate our findings regarding patients in the low CMMS group who responded better to immunotherapy.

Conclusions

CMMS detection has significant utility in the prognosis of patients at an early stage and identification of potential candidates for immunotherapy.