Building A Survival Prediction Model For Female Cancers By Integrating Multi-Omics Data

Ovarian cancer,cervical cancer,and endometrial cancer are the top three gynecologic malignancies with the highest incidence rate in women worldwide,which seriously threaten women’s health.Due to the heterogeneity of cancer,developing new methods based on multi-omics data for cancer molecular subtyping closely related to prognosis is of great practical significance for the prevention of recurrence and metastasis.In this paper,we use multi-omics data and deep learning and machine learning methods to study the subtypes of female cancers and their relationship with prognosis.The main work and research results are as follows:We downloaded DNA methylation,RPPA protein,mRNA expression,miRNA expression 4 multi-omics data and clinical data from TCGA database of ovarian cancer,cervical cancer,and endometrial cancer patients,and carried out data pre-processing.Using deep learning autoencoder method,we extracted the disease’s deep features and screened out the characteristics related to recurrence through univariate Cox proportional hazards model.We used K-means clustering algorithm to cluster samples and got the best AUC scoring model for three types of cancer,respectively.Then,using the survival-related deep features,we divided the three types of cancer patients into two subgroups,respectively.Finally,we used Kaplan-Meier methods and Log-rank test to detect the survival differences of the two subgroups of three cancers.Based on the genomic features associated with the two subgroups obtained by clustering,we used Lasso regression method to get the most relevant genomes,and further used Stacking ensemble learning to predict the two subgroups,and obtained a more concise and efficient model that can predict the death-related subgroups.In summary,this paper uses multi-omics data and deep learning and machine learning methods to explore cancer subtypes and survival analysis.The experimental results in three types of female cancer data in TCGA proved that the proposed clustering model can effectively discover different cancer subtypes with significant survival feature differences,realizes a new method of molecular subtyping based on survival sensitivity and multi-omics data,and establishes survival analysis models based on subtypes.Based on the analysis of the genomic differences,we further screened out different genes,which have reference significance for the subtyping and treatment of female cancers.