Fig. 1.

An overview of the proposed convolutional variational autoencoder (CVAE) model. This architecture includes two parts, an encoder and a decoder that are two symmetrical and reversed structures. Each one is composed of two convolutional layers and one dense layer. The latent 100-dimensional vector is the sampling layer (Z) generated using the mean and the standard deviation layers (i.e., μ and σ). We trained this architecture to take the input data, a stack of three integrated omics data (mRNA, microRNA, and DNA methylation), and automatically learn the latent vector’s distinguishing features.