Next generation sequencing (NGS) and single-cell sequencing technologies have enabled unparalleled resolution of the cell's molecular machinery; however, gleaning accurate knowledge from sequencing data is often stymied by high-dimensionality, measurement noise and biological complexity. Numerous methods have been proposed to address this by mapping high-dimensional inputs to informative low-dimensional representations. One such dimensionality reduction method is the variational autoencoder (VAE), which attempts to learn the probability distribution of given data through a low-dimensional latent variable, and has been shown to competitively separate cell types and to characterize functional differences of cell state1. In this talk, we present our preliminary results implementing a novel VAE model that incorporates prior knowledge through a VNN to create low-dimensional cell-state representations using bulk and single-cell RNA expression features. We hypothesize that this will reduce model complexity (number of parameters) while maintaining or improving model performance and creating a biologically relevant low-dimensional representation of sequencing data. Successful execution of this research will provide an interpretable and informative deep learning dimensionality reduction algorithm. This work was originally motivated by challenges in precision oncology, but may have application in many biological domains.