pyvae implements an Informed Variational Autoencoder (iVAE) whose encoder layers are constrained by biological priors (e.g. Reactome pathways) for interpretable representation learning on gene-expression data.