This repository contains an implementation of the multiGroupVI model described in "Disentangling shared and group-specific variations in single-cell transcriptomics data with multiGroupVI".
Overview of multiGroupVI. a, Given cells divided into non-overlapping groups of interest, multiGroupVI deconvolves the variations shared across groups versus those specific to individual groups. b, Schematic of the multiGroupVI model. A shared encoder network embeds cells, regardless of group membership, into the model's shared latent space, which captures variations shared across all groups. Group-specific encoders also embed cells into group-specific latent spaces, which capture variations particular to a given group. For a cell from a given group γ, the group-specific latent variables for other groups γ' ≠ γ are fixed to be zero vectors. Cells' latent representations are decoded back to the full gene expression space using a shared decoder. Here for simplicity we depict only two groups, though the model can natively handle more groups by adding additional group-specific encoders.
Git clone or download this repository. Then navigate to the folder and run
pip install .
A notebook demonstrating how to train a multiGroupVI model as well as for reproducing
our results on the scRNA-seq dataset considered in our paper can be found in
notebooks/Haber/create_haber_region_dataset.ipynb.
@article{multiGroupVI,
title={Disentangling shared and group-specific variations in single-cell transcriptomics data with multiGroupVI},
author={Weinberger, Ethan and Lopez, Romain and H\"utter, Jan-Christian and Regev, Aviv },
booktitle={Proceedings of the 17th Machine Learning in Computational Biology meeting},
year={2022},
organization={PMLR}
}