Deep Learning Approach for Breast Cancer InClust 5 Prediction based on Multiomics Data Integration

Breast cancer is the most common cancer among North American women and worldwide. In this paper, we present a deep learning model based on multiomics data integration to predict the five-year interval survival of breast cancer InClust 5. The data was selected from METABRIC dataset that contains three omic datasets: gene expression, copy number alteration (CNA), and clinical feature datasets. The model utilizes self-organizing map (SOM), which is an unsupervised method, to create an RGB to extract feature map for each omic to be the based for the convolution layer in the convolutional neural network CNN. In total, the model creates three CNN, one for each model. This method is the expansion of the iSOM-GSN model, where we create a feature map for each omic dataset instead of only one. The model incorporates the prediction of the three CNNs using an integration layer. The integration layer votes based on the prediction of the majority as the output of the model. The main contributions are 1) integrating multiomics data module, where the models learn from all the omic datasets. 2) a model to classify 1-a Dimensional sample vector using CNN. The results show high-performance measurement where the accuracy around 94 percent.