Discovery of Cancer Subtypes Based on Stacked Autoencoder

摘要

The discovery of cancer subtypes has become one of the research hotspots in bioinformatics. Clustering can be used to divide the same cancer into different subtypes, which can provide a basis and guidance for precision medicine and personalized medicine, so as to improve the treatment effect. It was found that multi-omics clustering had better effect than single cluster of omics data. However, omics data is usually of high dimensionality and noisy, and there are some challenges in multi-omics clustering. In this paper, we first use a stacked autoencoder neural network to reduce the dimensionality of multi-omics data and obtain the feature representation of low dimension. Then the similarity matrix is constructed by scaled exponential similarity kernel. Finally, we use spectral clustering method to calculate the clustering results. The experimental results on three datasets show that our method is more effective than the traditional dimensionality reduction method.