User

Import scvi-tools as:

import scvi

Model

model.AUTOZI	Automatic identification of zero-inflated genes [Clivio et al., 2019].
model.CondSCVI	Conditional version of single-cell Variational Inference, used for multi-resolution deconvolution of spatial transcriptomics data [Lopez et al., 2022].
model.DestVI	Multi-resolution deconvolution of Spatial Transcriptomics data (DestVI) [Lopez et al., 2022].
model.LinearSCVI	Linearly-decoded VAE [Svensson et al., 2020].
model.PEAKVI	Peak Variational Inference for chromatin accessilibity analysis [Ashuach et al., 2022].
model.SCANVI	Single-cell annotation using variational inference [Xu et al., 2021].
model.SCVI	single-cell Variational Inference [Lopez et al., 2018].
model.TOTALVI	total Variational Inference [Gayoso et al., 2021].
model.MULTIVI	Integration of multi-modal and single-modality data [Ashuach et al., 2023].
model.AmortizedLDA	Amortized Latent Dirichlet Allocation [Blei et al., 2003].
model.JaxSCVI	EXPERIMENTAL single-cell Variational Inference [Lopez et al., 2018], but with a Jax backend.

External models

external.CellAssign	Reimplementation of CellAssign for reference-based annotation [Zhang et al., 2019].
external.GIMVI	Joint VAE for imputing missing genes in spatial data [Lopez et al., 2019].
external.RNAStereoscope	Reimplementation of Stereoscope [Andersson et al., 2020] for deconvolution of spatial transcriptomics from single-cell transcriptomics.
external.SpatialStereoscope	Reimplementation of Stereoscope [Andersson et al., 2020] for deconvolution of spatial transcriptomics from single-cell transcriptomics.
external.SOLO	Doublet detection in scRNA-seq [Bernstein et al., 2020].
external.SCAR	Ambient RNA removal in scRNA-seq data [Sheng et al., 2022].
external.Tangram	Reimplementation of Tangram [Biancalani et al., 2021] for mapping single-cell RNA-seq data to spatial data.
external.SCBASSET	EXPERIMENTAL Reimplementation of scBasset [Yuan and Kelley, 2022] for representation learning of scATAC-seq data.
external.ContrastiveVI	contrastive variational inference [Weinberger et al., 2023].
external.POISSONVI	Peak Variational Inference using a Poisson distribution [Martens et al., 2023].

Data loading

scvi-tools relies entirely on the AnnData format. For convenience, we have included data loaders from the AnnData API. Scanpy also has utilities to load data that are outputted by 10x’s Cell Ranger software.

data.read_h5ad	Read .h5ad-formatted hdf5 file.
data.read_csv	Read .csv file.
data.read_loom	Read .loom-formatted hdf5 file.
data.read_text	Read .txt, .tab, .data (text) file.
data.read_10x_atac	Read scATAC-seq data outputted by 10x Genomics software.
data.read_10x_multiome	Read Multiome (scRNA + scATAC) data outputted by 10x Genomics software.

Basic preprocessing

For general single-cell preprocessing, we defer to our friends at Scanpy, and specifically their preprocessing module (scanpy.pp).

All scvi-tools models require raw UMI count data. The count data can be safely stored in an AnnData layer as one of the first steps of a Scanpy single-cell workflow:

adata.layers["counts"] = adata.X.copy()

Here we maintain a few package specific utilities for feature selection, etc.

data.poisson_gene_selection	Rank and select genes based on the enrichment of zero counts.
data.organize_cite_seq_10x	Organize anndata object loaded from 10x for scvi models.
data.organize_multiome_anndatas	Concatenate multiome and single-modality input anndata objects.
data.add_dna_sequence	Add DNA sequence to AnnData object.
data.reads_to_fragments	Convert scATAC-seq read counts to appoximate fragment counts.

Model hyperparameter tuning

scvi-tools supports automatic model hyperparameter tuning using Ray Tune.

autotune.ModelTuner	Automated and scalable hyperparameter tuning for scvi-tools models.
autotune.TuneAnalysis	Dataclass for storing results from a tuning experiment.

Model hub

We have a hub for pre-trained scvi-tools models that is hosted on huggingface. Using the functionality that scvi-tools provides, users can download pre-trained scvi-tools models (and datasets) from this platform, and model generators can upload their own pre-trained scvi-tools models to this platform.

hub.HubMetadata	Encapsulates the required metadata for scvi-tools hub models.
hub.HubModelCardHelper	A helper for creating a ModelCard for scvi-tools hub models.
hub.HubModel	Provides functionality to interact with the scvi-hub backed by huggingface.

Model criticism

criticism.PosteriorPredictiveCheck

EXPERIMENTAL Posterior predictive checks for comparing scRNA-seq generative models.

Utilities

Here we maintain miscellaneous general methods.

model.utils.mde	Util to run pymde.preserve_neighbors() for visualization of scvi-tools embeddings.
model.utils.get_minified_adata_scrna	Returns a minified adata that works for most scrna models (such as SCVI, SCANVI).

Configuration

An instance of the ScviConfig is available as scvi.settings and allows configuring scvi-tools.

_settings.ScviConfig

Config manager for scvi-tools.