Novel Gene Signatures Predicting Breast Cancer Based on Random Forest and Artificial Neural Network

Background: Breast cancer is the most common malignancy in women worldwide, which seriously threatens women's physical and mental health, but currently, there is no classification method for tumor samples based on gene expression profiles for faster breast cancer diagnosis. The study aimed to establish a novel genetic model to distinguish breast cancer patients from the normal population. Results: We utilized published expression profiles of breast cancer patients (GSE15852, GSE70905) to identify potential predictive gene panels. A total of 7 differentially expressed genes were identified as predictors. Random forest algorithm and artificial neural network were applied to screen the predictive features and build a model to predict breast cancer. In parallel, we validated this prediction model using expression profiling of a completely independent set of breast cancer patients(GSE70947). The new model was successfully built based on the molecular prognostic scoring system and showed significant predictive value in the training group (AUC = 0.991), which was simultaneously validated in an independent dataset (AUC = 0.817). Conclusions: Random forest algorithm combined with artificial neural network successfully constructed a prediction model for breast cancer. The new model can predict breast cancer patients, which is helpful for the diagnosis of breast cancer in the clinic.