Gene Feature Extraction and Cancer Classification via Double Network

Gene expression profiles are widely used in disease discovery and drug efficacy analysis, which can describe gene expression types and abundance information in specific cells or tissues under different conditions. Faced with a large number of high-dimensional and complex gene expression data, it has become a research topic that how to extract key information and use it effectively in the field of bioinformatics. Based on different types of gene expression profiles from GEO and TCGA database, the double network structure via deep learning is designed in this paper, which is applied for gene feature extraction and cancer classification. The feature extraction method based on double network both considers the classification ability and reconstruction ability of extracted features, which can be restored to original gene expression. According to requirements of sample classification and feature extraction, the loss function of training process is redefined, which implements better gene reconstruction and model training. And massive experiments show that double network extracts more precise gene features and improves accuracy of cancer classification.