Dr. Moritz Hess
—
abgelegt unter:
FDM-Seminar
Deep generative approaches for omics data: interpretability and sample-size constraints
Deep generative models (DGMs) are promising tools, e.g., for learning latent structure in high dimensional omics data such as single-cell RNA-Seq data as well as for generating synthetic observations, e.g. for securely sharing single nucleotide polymorphism (SNP) data. Here I address the interpretability of DGMs, specifically by showing how to link latent space information with observed variables (e.g. expression levels of genes). In addition I address the performance of DGMs under sample size constraints which are frequently observable when working with omics data in the biomedical context.