Scvi-tools

Major changes

- Model hyperparameter tuning is available through {class}`~scvi.autotune.ModelTuner` (beta)
{pr}`1785`,{pr}`1802`,{pr}`1831`.
- Pre-trained models can now be uploaded to and downloaded from [Hugging Face models] using the
{mod}`~scvi.hub` module {pr}`1779`,{pr}`1812`,{pr}`1828`,{pr}`1841`, {pr}`1851`,{pr}`1862`.
- {class}`~anndata.AnnData` `.var` and `.varm` attributes can now be registered through new fields
in {mod}`~scvi.data.fields` {pr}`1830`,{pr}`1839`.
- {class}`~scvi.external.SCBASSET`, a reimplementation of the [original scBasset model], is
available for representation learning of scATAC-seq data (experimental) {pr}`1839`,{pr}`1844`,
{pr}`1867`,{pr}`1874`,{pr}`1882`.
- {class}`~scvi.train.LowLevelPyroTrainingPlan` and {class}`~scvi.model.base.PyroModelGuideWarmup`
added to allow the use of vanilla PyTorch optimization on Pyro models {pr}`1845`,{pr}`1847`.
- Add {meth}`scvi.data.cellxgene` function to download cellxgene datasets {pr}`1880`.

Minor changes

- Latent mode support changed so that user data is no longer edited in-place {pr}`1756`.
- Minimum supported PyTorch Lightning version is now 1.9 {pr}`1795`,{pr}`1833`,{pr}`1863`.
- Minimum supported Python version is now 3.8 {pr}`1819`.
- [Poetry] removed in favor of [Hatch] for builds and publishing {pr}`1823`.
- `setup_anndata` docstrings fixed, `setup_mudata` docstrings added {pr}`1834`,{pr}`1837`.
- {meth}`~scvi.data.add_dna_sequence` adds DNA sequences to {class}`~anndata.AnnData` objects using
[genomepy] {pr}`1839`,{pr}`1842`.
- Update tutorial formatting with pre-commit {pr}`1850`
- Expose `accelerators` and `devices` arguments in {class}`~scvi.train.Trainer` {pr}`1864`.
- Development in GitHub Codespaces is now supported {pr}`1836`.

Breaking changes

- {class}`~scvi.module.base.LossRecorder` has been removed in favor of
{class}`~scvi.module.base.LossOutput` {pr}`1869`.

Bug Fixes

- {class}`~scvi.train.JaxTrainingPlan` now correctly updates `global_step` through PyTorch
Lightning by using a dummy optimizer. {pr}`1791`.
- CUDA compatibility issue fixed in {meth}`~scvi.distributions.ZeroInflatedNegativeBinomial.sample`
{pr}`1813`.
- Device-backed {class}`~scvi.dataloaders.AnnTorchDataset` fixed to work with sparse data {pr}`1824`.
- Fix bug {meth}`~scvi.model.base._log_likelihood.compute_reconstruction_error` causing the first
batch to be ignored, see more details in {issue}`1854` {pr}`1857`.

Contributors

- {ghuser}`adamgayoso`
- {ghuser}`eroell`
- {ghuser}`gokceneraslan`
- {ghuser}`macwiatrak`
- {ghuser}`martinkim0`
- {ghuser}`saroudant`
- {ghuser}`vitkl`
- {ghuser}`watiss`

Major Changes

- {class}`~scvi.train.TrainingPlan` allows custom PyTorch optimizers [1747].
- Improvements to {class}`~scvi.train.JaxTrainingPlan` [1747] [1749].
- {class}`~scvi.module.base.LossRecorder` is deprecated. Please substitute with
{class}`~scvi.module.base.LossOutput` [1749]
- All training plans require keyword args after the first positional argument [1749]
- {class}`~scvi.module.base.JaxBaseModuleClass` absorbed features from the `JaxModuleWrapper`,
rendering the `JaxModuleWrapper` obsolote, so it was removed. [1751]
- Add {class}`scvi.external.Tangram` and {class}`scvi.external.tangram.TangramMapper` that
implement Tangram for mapping scRNA-seq data to spatial data [1743].

Minor changes

- Remove confusing warning about kl warmup, log kl weight instead [1773]

Breaking changes

- {class}`~scvi.module.base.LossRecorder` no longer allows access to dictionaries of values if
provided during initialization [1749].
- `JaxModuleWrapper` removed. [1751]

Bug Fixes

- Fix `n_proteins` usage in {class}`~scvi.model.MULTIVI` [1737].
- Remove unused param in {class}`~scvi.model.MULTIVI` [1741].
- Fix random seed handling for Jax models [1751].

Contributors

- [watiss]
- [adamgayoso]
- [martinkim0]
- [marianogabitto]

Major Changes

- Add latent mode support in {class}`~scvi.model.SCVI` [1672]. This allows for loading a model
using latent representations only (i.e. without the full counts). Not only does this speed up
inference by using the cached latent distribution parameters (thus skipping the encoding step),
but this also helps in scenarios where the full counts are not available but cached latent
parameters are. We provide utility functions and methods to dynamically convert a model to
latent mode.
- Added {class}`~scvi.external.SCAR` as an external model for ambient RNA removal [1683].

Minor changes

- Faster inference in PyTorch with `torch.inference_mode` [1695].
- Upgrade to Lightning 1.6 [1719].
- Update CI workflow to separate static code checking from pytest [1710].
- Add Python 3.10 to CI workflow [1711].
- Add {meth}`~scvi.data.AnnDataManager.register_new_fields` [1689].
- Use sphinxcontrib-bibtex for references [1731].
- {meth}`~scvi.model.base.VAEMixin.get_latent_representation`: more explicit and better docstring
[1732].
- Replace custom attrdict with {class}`~ml_collections` implementation [1696].

Breaking changes

- Add weight support to {class}`~scvi.model.MULTIVI` [1697]. Old models can't be loaded anymore.

Bug Fixes

- Fix links for breast cancer and mouse datasets [1709].
- fix quick start notebooks not showing [1733].

Contributors

- [watiss]
- [adamgayoso]
- [martinkim0]
- [ricomnl]
- [marianogabitto]

Changes

- Support for PyTorch Lightning 1.7 [1622].
- Allow `flax` to use any mutable states used by a model generically with
{class}`~scvi.module.base.TrainStateWithState` [1665], [1700].
- Update publication links in `README` [1667].
- Docs now include floating window cross references with `hoverxref`, external links with
`linkcode`, and `grid` [1678].

Bug Fixes

- Fix `get_likelihood_parameters()` failure when `gene_likelihood != "zinb"` in
{class}`~scvi.model.base.RNASeqMixin` [1618].
- Fix exception logic when not using the observed library size in {class}`~scvi.module.VAE`
initialization [1660].
- Replace instances of `super().__init__()` with an argument in `super()`, causing `autoreload`
extension to throw errors [1671].
- Change cell2location tutorial causing docs build to fail [1674].
- Replace instances of `max_epochs` as `int`s for new PyTorch Lightning [1686].
- Catch case when `torch.backends.mps` is not implemented [1692].
- Fix Poisson sampling in {meth}`~scvi.module.VAE.sample` [1702].

Contributors

- [adamgayoso]
- [watiss]
- [mkarikom]
- [tommycelsius]
- [ricomnl]