The Compositional Perturbation Autoencoder (CPA) is a deep generative framework to learn effects of perturbations at the single-cell level. CPA performs OOD predictions of unseen combinations of drugs, learns interpretable embeddings, estimates dose-response curves, and provides uncertainty estimates.
Please see https://github.com/igraph/python-igraph/issues/699 for the explanation and please consider making this change in other theislab projects as well, if possible. Ref: https://github.com/search?q=org%3Atheislab%20python-igraph&type=code