Neural network has shown great success in analyzing imaging data and natural language processing. Recently, it has also demonstrated good performance in some genomics problems. Here we two examples of analyzing genomic data using neural networks. One is to combining gene expression as well as functional annotation (e.g., gene-gene interaction or gene ontology) to identify meaningful gene expression features, by two types of neuron networks, auto-encoder and graph neural network. The other example is to analyze somatic mutation data using neural network. The somatic mutation data are highly sparse (e.g., one gene may be mutated in only one or a few individuals) and thus it is challenging to analyze using traditional method. We show that neural network method can identify meaningful latent structures from somatic mutation data.