Clemens Schächter (AG Prof. Dr. Harald Binder: Knowledge Discovery Synthesis)  - Enhancement of network architecture alignment in comparative single-cell studies

Abstract

"Animal models can provide meaningful context for human single-cell data. To transfer information between species, we propose a deep learning approach that pretrains a conditional variational autoencoder on animal data and reuses its last encoder layers in the human network architecture during fine-tuning. This unifies latent spaces and enables information transfer across species, even when gene sets differ. When applied to cross-species pairs of liver, adipose tissue, and glioblastoma datasets, our method learns an aligned representation that can be used for label transfer. Thus, we reliably uncover and exploit similarities between species to provide context for human single-cell data."