Essential proteins are group of proteins that are indispensable to survival and development of cells. Prediction and analysis of essential genes/proteins are crucial for uncovering the mechanisms of cells. Using bioinformatics and high-throughput technologies, forecasting essential genes/proteins by protein-protein interaction (PPI) networks have become more efficient than traditional approaches which use expensive and time-consuming experimental methods. Previous studies have found that the essentiality of genes closely relates to their properties in PPI network. In this work, we propose a supervised deep model for predicting human essential genes using neighboring details of genes/proteins in the PPI network. Our approach implements a weight-biased random walk on PPI network to get the node network context. Then, some different measures are used to get some feature vectors for each node (gene/protein) that preserve the network structure as well as the gene's properties in the PPI network. These feature vectors are then fed to a Relational AutoEncoder to embed the genes' features into latent space. At last, these embedded features are put into a trained classifier to predict the human essential genes. The prediction results on two human PPI networks show that our model achieves better performance than those that only refer to genes' centrality properties in the network.