mrvi 1.0.1

Multi-resolution Variational Inference

mrvi

Multi-resolution Variational Inference.

🚧 :warning: [DEPRECATED] For the latest version of MrVI, please use the implementation in the scvi-tools package (from scvi.external import MRVI). This package will no longer be maintained. :warning: 🚧

Getting started

Please refer to the documentation. In particular, the

• API documentation.

Installation

You need to have Python 3.8 or newer installed on your system. If you don't have Python installed, we recommend installing Miniconda.

There are several alternative options to install mrvi:

1. Install the latest pre-release of mrvi from PyPI <https://pypi.org/project/mrvi/>_:

pip install --pre mrvi

2. Install the latest development version:

pip install git+https://github.com/YosefLab/mrvi.git@main

User guide

While a more comprehensive user guide is in the works, you can find here a brief overview of the main features of mrvi.

Data preparation: MrVI relies on scvi-tools routines for model initialization and training. In particular, mrvi assumes data to be stored in an AnnData object. A first step is to load the data and register it, as follows:

from mrvi import MrVI

MrVI.setup_anndata(adata, sample_key="my_sample_key", batch_key="my_batch_key")

where here 'my_sample_key' and 'my_batch_key' are expected to be keys of adata.obs that contain the sample and batch assignments, respectively.

Model training: The next step is to initialize and train the model, which can be done via:

model = MrVI(adata)
model.train()

Once the model is trained, we recommend visualizing the validation ELBO to assess convergence, which is stored in model.history["elbo_validation"]. If the ELBO has not converged, you should consider training the model for more epochs.

Latent space visualization: MrVI contains two latent spaces, u, that captures global cell-type variations, and z, that additionally captures sample-specific variations. These two latent representations can be accessed via model.get_latent_representation(), (with give_z=True to access z). In particular, these latent variables can be seemlessly used for data visualization or clustering using scanpy. For instance, visualizing the u latent space can be done via:

import scanpy as sc
from scvi.model.utils import mde

u = model.get_latent_representation()
u_mde = mde(u)
adata.obsm["u_mde"] = u_mde
sc.pl.embedding(adata, basis="u_mde")

Computing sample-sample dissimilarities: MrVI can be used to predict sample-sample dissimilarities, using the following snippet:

# Predict sample-sample dissimilarities per cell type
dists = model.get_local_sample_distances(
    adata, keep_cell=False, groupby="initial_clustering", batch_size=32
)

# OR predict sample-sample dissimilarities for EACH cell
# WARNING: this can be slow and memory-intensive for large datasets
dists = model.get_local_sample_distances(adata, keep_cell=True, batch_size=32)

These dissimilarities can then be visualized via seaborn.clustermap or similar tools.

DE analysis: MrVI can be used to identify differentially expressed genes (DEGs) between two groups of samples at the single-cell level. Here, "samples" refere to the sample_key provided in MrVI.setup_anndata. Identifying such genes can be done as follows,

sample_cov_keys = ["Status"]  # Here, Status is the sample covariate of interest
de_res = model.differential_expression(
    sample_cov_keys=sample_cov_keys,
)

DA analysis: MrVI can also be used to assess differences in cell-type compositions across sample groups, using the following snippet:

da_res = model.differential_abundance(sample_cov_keys=sample_cov_keys)
A_log_probs = da_res.Status_log_probs.loc[{"Status": "A"}]
B_log_probs = da_res.Status_log_probs.loc[{"Status": "B"}]
A_B_log_prob_ratio = A_log_probs - B_log_probs

Release notes

See the changelog.

Contact

For questions and help requests, you can reach out in the scverse discourse. If you found a bug, please use the issue tracker.

Citation

Deep generative modeling of sample-level heterogeneity in single-cell genomics

Pierre Boyeau, Justin Hong, Adam Gayoso, Martin Kim, Jose L. McFaline-Figueroa, Michael I. Jordan, Elham Azizi, Can Ergen, & Nir Yosef

bioRxiv 2024 May 10. doi: 10.1101/2022.10.04.510898.