scDRP documentation

This is a package to analyze single-cell perturbation data. It learns the latent factors that generates the observed gene profiles via sparsity-regularized disentangled VAE, which disentangles the latent space into perturbation-dependent and perturbation-invariant subspaces. Leveraging the disentangled latent factors, we can estimate individual treatment effects (ITE) and generate counterfactual samples via soft-conditional optimal transport. This is based on the idea that the effect of perturbation on those perturbation-dependent latent factors should be rank-preserved when conditioning on confounders.

Contents:

• scDRP
  • scDRP package