Developer

Import scvi-tools as:

import scvi

Data Registration

AnnDataFields delineate how scvi-tools refers to fields in AnnData objects. The AnnDataManager provides an interface for operating over a collection of AnnDataFields and an AnnData object.

data.AnnDataManager	Provides an interface to validate and process an AnnData object for use in scvi-tools.
data.AnnDataManagerValidationCheck	Validation checks for AnnOrMudata scvi-tools compatibility.
data.fields.BaseAnnDataField	Abstract class for a single AnnData/MuData field.
data.fields.LayerField	An AnnDataField for layer or X attributes in the AnnData data structure.
data.fields.CategoricalObsField	An AnnDataField for categorical .obs attributes in the AnnData data structure.
data.fields.CategoricalVarField	An AnnDataField for categorical .var attributes in the AnnData data structure.
data.fields.NumericalJointObsField	An AnnDataField for a collection of numerical .obs fields in the AnnData data structure.
data.fields.NumericalJointVarField	An AnnDataField for a collection of numerical .var fields in the AnnData data structure.
data.fields.CategoricalJointObsField	An AnnDataField for a collection of categorical .obs fields in AnnData.
data.fields.CategoricalJointVarField	An AnnDataField for a collection of categorical .var fields in AnnData.
data.fields.ObsmField	An AnnDataField for an .obsm field in the AnnData data structure.
data.fields.VarmField	An AnnDataField for a .varm field in the AnnData data structure.
data.fields.ProteinObsmField	An AnnDataField for an protein data stored in an .obsm field of an AnnData object.
data.fields.StringUnsField	An AnnDataField for string .uns attributes in the AnnData data structure.
data.fields.LabelsWithUnlabeledObsField	An AnnDataField for labels which include explicitly unlabeled cells.
data.fields.BaseMuDataWrapperClass	A wrapper class that adds MuData support for an AnnDataField.
data.fields.MuDataWrapper	Wraps an AnnDataField with BaseMuDataWrapperClass.
data.fields.MuDataLayerField
data.fields.MuDataProteinLayerField
data.fields.MuDataNumericalObsField
data.fields.MuDataNumericalVarField
data.fields.MuDataCategoricalObsField
data.fields.MuDataCategoricalVarField
data.fields.MuDataObsmField
data.fields.MuDataVarmField
data.fields.MuDataNumericalJointObsField
data.fields.MuDataNumericalJointVarField
data.fields.MuDataCategoricalJointObsField
data.fields.MuDataCategoricalJointVarField
data.AnnTorchDataset	Extension of Dataset for AnnData objects.

Data Loaders

DataLoaders for loading tensors from AnnData objects. DataSplitters for splitting data into train/test/val.

dataloaders.AnnDataLoader	DataLoader for loading tensors from AnnData objects.
dataloaders.AnnTorchDataset	Extension of Dataset for AnnData objects.
dataloaders.CollectionAdapter	Allow an AnnCollection to pretend to be an AnnData in SCVI-Tools
dataloaders.ConcatDataLoader	DataLoader that supports a list of list of indices to load.
dataloaders.DataSplitter	Creates data loaders train_set, validation_set, test_set.
dataloaders.SemiSupervisedDataLoader	DataLoader that supports semisupervised training.
dataloaders.SemiSupervisedDataSplitter	Creates data loaders train_set, validation_set, test_set.
dataloaders.BatchDistributedSampler	EXPERIMENTAL Sampler that restricts to loading from a subset of the dataset.
dataloaders.MappedCollectionDataModule
dataloaders.TileDBDataModule	PyTorch Lightning DataModule for training scVI models from SOMA data

Distributions

Parameterizable probability distributions.

distributions.Poisson	Poisson distribution.
distributions.NegativeBinomial	Negative binomial distribution.
distributions.NegativeBinomialMixture	Negative binomial mixture distribution.
distributions.ZeroInflatedNegativeBinomial	Zero-inflated negative binomial distribution.
distributions.JaxNegativeBinomialMeanDisp	Negative binomial parameterized by mean and inverse dispersion.
distributions.BetaBinomial	Beta binomial distribution.

Model (Base)

These classes should be used to construct user-facing model classes.

model.base.BaseModelClass	Abstract class for scvi-tools models.
model.base.BaseMinifiedModeModelClass	Abstract base class for scvi-tools models that can handle minified data.
model.base.VAEMixin	Universal variational auto-encoder (VAE) methods.
model.base.RNASeqMixin	General purpose methods for RNA-seq analysis.
model.base.ArchesMixin	Universal scArches implementation.
model.base.UnsupervisedTrainingMixin	General purpose unsupervised train method.
model.base.SemisupervisedTrainingMixin	General purpose semisupervised train, predict, and interoperability methods.
model.base.PyroSviTrainMixin	Mixin class for training Pyro models.
model.base.PyroSampleMixin	Mixin class for generating samples from posterior distribution.
model.base.PyroJitGuideWarmup	A callback to warmup a Pyro guide.
model.base.PyroModelGuideWarmup	A callback to warmup a Pyro guide and model.
model.base.DifferentialComputation	Unified class for differential computation.
model.base.EmbeddingMixin	EXPERIMENTAL Mixin class for initializing and using embeddings in a model.

Module

Existing module classes with respective generative and inference procedures.

module.AutoZIVAE	Implementation of the AutoZI model [Clivio et al., 2019].
module.Classifier	Basic fully-connected NN classifier.
module.LDVAE	Linear-decoded Variational auto-encoder model.
module.MRDeconv	Model for multi-resolution deconvolution of spatial transriptomics.
module.PEAKVAE	Variational auto-encoder model for ATAC-seq data.
module.MULTIVAE	Variational auto-encoder model for joint paired + unpaired RNA-seq and ATAC-seq data.
module.SCANVAE	Single-cell annotation using variational inference.
module.TOTALVAE	Total variational inference for CITE-seq data.
module.VAE	Variational auto-encoder [Lopez et al., 2018].
module.VAEC	Conditional Variational auto-encoder model.
module.AmortizedLDAPyroModule	An amortized implementation of Latent Dirichlet Allocation [Blei et al., 2003].
module.JaxVAE	Variational autoencoder model.

External module

Module classes in the external API with respective generative and inference procedures.

external.gimvi.JVAE	Joint variational auto-encoder for imputing missing genes in spatial data.
external.cytovi.CytoVAE	Variational auto-encoder model for Cytometry.
external.cellassign.CellAssignModule	Model for CellAssign.
external.contrastivevi.ContrastiveDataSplitter	Creates ContrastiveDataLoader for training, validation, and test set.
external.stereoscope.RNADeconv	Model of scRNA-seq for deconvolution of spatial transcriptomics.
external.stereoscope.SpatialDeconv	Model of single-cell RNA-sequencing data for deconvolution of spatial transcriptomics.
external.tangram.TangramMapper	Tangram Mapper Model.
external.scbasset.ScBassetModule	PyTorch implementation of ScBasset [Yuan and Kelley, 2022].
external.contrastivevi.ContrastiveVAE	Variational inference for contrastive analysis of RNA-seq data.
external.velovi.VELOVAE	Variational auto-encoder model.
external.mrvi_jax.JaxMRVAE	Multi-resolution Variational Inference (MrVI) module.
external.mrvi_torch.TorchMRVAE	Multi-resolution Variational Inference (MrVI) module.
external.methylvi.METHYLVAE	PyTorch module for methylVI.
external.methylvi.METHYLANVAE	Methylation annotation using variational inference.
external.decipher.DecipherPyroModule	Pyro Module for the Decipher model.
external.resolvi.RESOLVAE	Implementation of resolVI.
external.scviva.nicheVAE	Variational auto-encoder with niche decoders [Levy et al., 2025].
external.scviva.NicheLossOutput	Modify loss output to record niche losses.
external.sysvi.SysVAE	CVAE with optional VampPrior and latent cycle consistency loss.

Module (Base)

These classes should be used to construct module classes that define generative models and inference schemes.

module.base.BaseModuleClass	Abstract class for scvi-tools modules.
module.base.BaseMinifiedModeModuleClass	Abstract base class for scvi-tools modules that can handle minified data.
module.base.SupervisedModuleClass	General purpose supervised classify and loss calculations methods.
module.base.PyroBaseModuleClass	Base module class for Pyro models.
module.base.JaxBaseModuleClass	Abstract class for Jax-based scvi-tools modules.
module.base.EmbeddingModuleMixin	EXPERIMENTAL Mixin class for initializing and using embeddings in a module.
module.base.LossOutput	Loss signature for models.
module.base.auto_move_data	Decorator for Module methods to move data to correct device.

Neural networks

Basic neural network building blocks.

nn.FCLayers	A helper class to build fully-connected layers for a neural network.
nn.Encoder	Encode data of n_input dimensions into a latent space of n_output dimensions.
nn.Decoder	Decodes data from latent space to data space.
nn.DecoderSCVI	Decodes data from a latent space of n_input dimensions into n_output dimensions.
nn.LinearDecoderSCVI	Linear decoder for scVI.
nn.Embedding	EXPERIMENTAL Embedding layer with utility methods for extending.
nn.DecoderTOTALVI	Decodes data from a latent space of n_input dimensions n_output dimensions.
nn.EncoderTOTALVI	Encodes data of n_input dimensions into a latent space of n_output dimensions.

Train

TrainingPlans define train/test/val optimization steps for modules.

train.AdversarialTrainingPlan	Train vaes with adversarial loss option to encourage latent space mixing.
train.ClassifierTrainingPlan	Lightning module task to train a simple MLP classifier.
train.SemiSupervisedTrainingPlan	Lightning module task for SemiSupervised Training.
train.SemiSupervisedAdversarialTrainingPlan	Lightning module task for SemiSupervised Training with Adversarial Loss.
train.LowLevelPyroTrainingPlan	Lightning module task to train Pyro scvi-tools modules.
train.PyroTrainingPlan	Lightning module task to train Pyro scvi-tools modules.
train.JaxTrainingPlan	Lightning module task to train Jax scvi-tools modules.
train.Trainer	Lightweight wrapper of Pytorch Lightning Trainer.
train.TrainingPlan	Lightning module task to train scvi-tools modules.
train.TrainRunner	TrainRunner calls Trainer.fit() and handles pre and post training procedures.
train.ScibCallback	A callback to initialize the Scib-Metrics autotune module.
train.SaveCheckpoint	Saves model checkpoints based on a monitored metric.
train.LoudEarlyStopping	Loud early stopping callback.

Utilities

Utility functions used by scvi-tools.

utils.track	Progress bar with 'rich' and 'tqdm' styles.
utils.setup_anndata_dsp	Class that is intended to process docstrings.
utils.attrdict	A dictionary that allows for attribute-style access.
external.decipher.utils.Trajectory	A class that defines a trajectory through a latent space.