Machine learning-based construction of a miRNAs prognostic risk assessment model of cervical cancer

Molecular profiling of cervical cancer are expected to promote clinical treatment of cervical cancer, but prognostic risk assessment models that clinically-integrated microRNAs (miRNAs) expression profiling data of cervical cancer have not been systematically studied. First, a random forest (RF) model was used to select seed miRNAs related with survival. In order to improve the survival prediction ability, the clinical variables of cervical cancer patients were integrated with miRNAs expression profile data, and the key miRNAs related with survival were selected from the seed miRNAs by bidirectional elimination model. Based on the machine learning model and clinically-integrated miRNAs expression profiling data, eight miRNAs biomarkers associated with cervical cancer prognosis were identified. Among these eight key miRNAs, miR-150 and miR-502 have been reported to be associated with cervical cancer in previous experimental studies, while the relationship between the remaining 6 miRNAs and cervical cancer has not been reported in literature. Then miRNAs prognostic risk assessment model of cervical cancer was successfully constructed. The results of the Kaplan-Meier survival curve showed that patients with high risk groups had a poorer prognosis (P<0.001), the area under curve (AUC) of the model 5-year overall survival rate was 0.94. Finally, compared with the the traditional cox proportional-hazards model (cox model) and the use of standalone miRNAs expression profile data and the model has good prognostic performance for cervical cancer patients.