Enhancing Breast Cancer Survival Prognosis Through Omic and Non-Omic Data Integration

Cancer, the second leading cause of death globally, claimed 685,000 lives among 2.3 million women affected by breast cancer in 2020. Cancer prognosis plays a pivotal role in tailoring treatments and assessing efficacy, emphasizing the need for a comprehensive understanding. The goal is to develop predictive model capable of accurately predicting patient outcomes and guiding personalized treatment strategies, thereby advancing precision medicine in breast cancer care. This project addresses limitations in current cancer prognosis models by integrating omics and non-omics data. While existing models often neglect crucial omics data like DNA methylation and miRNA, the method utilizes the TCGA dataset to incorporate these data types along with others. Employing mRMR feature selection and CNN models for each type of data for feature extraction, features are stacked and a Random Forest classifier is employed for final prognosis. The proposed method is applied to the dataset to predict whether the patient is a long-time or a short-time survivor. This strategy showcases excellent performance, with an AUC value of 0.873, precision at 0.881, and sensitivity reaching 0.943. With an accuracy rate of 0.861, signaling an improvement of 11.96% compared to prior studies. In conclusion, integrating diverse data with advanced machine learning holds promise for improving breast cancer prognosis. Addressing model limitations and leveraging comprehensive datasets can enhance accuracy, paving the way for better patient care. Further refinement offers potential for significant advancements in cancer prognosis and treatment strategies.