Scvi-tools

This release adds features for tighter integration with Pyro for model development, fixes for
{class}`~scvi.external.SOLO`, and other enhancements. Users of {class}`~scvi.external.SOLO` are
strongly encouraged to upgrade as previous bugs will affect performance.

Enchancements

- Add {class}`scvi.model.base.PyroSampleMixin` for easier posterior sampling with Pyro ([1059]).
- Add {class}`scvi.model.base.PyroSviTrainMixin` for automated training of Pyro models ([1059]).
- Ability to pass kwargs to {class}`~scvi.module.Classifier` when using
{class}`~scvi.external.SOLO` ([1078]).
- Ability to get doublet predictions for simulated doublets in {class}`~scvi.external.SOLO`
([1076]).
- Add "comparison" column to differential expression results ([1074]).
- Clarify {class}`~scvi.external.CellAssign` size factor usage. See class docstring.

Changes

- Update minimum Python version to `3.7.2` ([1082]).
- Slight interface changes to {class}`~scvi.train.PyroTrainingPlan`. `"elbo_train"` and
`"elbo_test"` are now the average over minibatches as ELBO should be on scale of full data and
`optim_kwargs` can be set on initialization of training plan ([1059], [1101]).
- Use pandas read pickle function for pbmc dataset metadata loading ([1099]).
- Adds `n_samples_overall` parameter to functions for denoised expression/accesibility/etc. This is
used in during differential expression ([1090]).
- Ignore configure optimizers warning when training Pyro-based models ([1064]).

Bug fixes

- Fix scale of library size for simulated doublets and expression in {class}`~scvi.external.SOLO`
when using observed library size to train original {class}`~scvi.model.SCVI` model ([1078],
[1085]). Currently, library sizes in this case are not appropriately put on the log scale.
- Fix issue where anndata setup with a layer led to errors in {class}`~scvi.external.SOLO`
([1098]).
- Fix `adata` parameter of {func}`scvi.external.SOLO.from_scvi_model`, which previously did nothing
([1078]).
- Fix default `max_epochs` of {class}`~scvi.model.SCANVI` when initializing using pre-trained model
of {class}`~scvi.model.SCVI` ([1079]).
- Fix bug in `predict()` function of {class}`~scvi.model.SCANVI`, which only occurred for soft
predictions ([1100]).

Breaking changes

None!

Contributors

- [vitkl]
- [adamgayoso]
- [galenxing]
- [PierreBoyeau]
- [Munfred]
- [njbernstein]
- [mjayasur]

From the user perspective, this release features the new differential expression functionality (to
be described in a manuscript). For now, it is accessible from
{func}`~scvi.model.SCVI.differential_expression`. From the developer perspective, we made changes
with respect to {class}`scvi.dataloaders.DataSplitter` and surrounding the Pyro backend. Finally,
we also made changes to adapt our code to PyTorch Lightning version 1.3.

Changes

- Pass `n_labels` to {class}`~scvi.module.VAE` from {class}`~scvi.model.SCVI` ([1055]).
- Require PyTorch lightning > 1.3, add relevant fixes ([1054]).
- Add DestVI reference ([1060]).
- Add PeakVI links to README ([1046]).
- Automatic delta and eps computation in differential expression ([1043]).
- Allow doublet ratio parameter to be changed for used in SOLO ([1066]).

Bug fixes

- Fix an issue where `transform_batch` options in {class}`~scvi.model.TOTALVI` was accidentally
altering the batch encoding in the encoder, which leads to poor results ([1072]). This bug was
introduced in version 0.9.0.

Breaking changes

These breaking changes do not affect the user API; though will impact model developers.

- Use PyTorch Lightning data modules for {class}`scvi.dataloaders.DataSplitter` ([1061]). This
induces a breaking change in the way the data splitter is used. It is no longer callable and
now has a `setup` method. See {class}`~scvi.train.TrainRunner` and its source code, which is
straightforward.
- No longer require training plans to be initialized with `n_obs_training` argument ([1061]).
`n_obs_training` is now a property that can be set before actual training to rescale the loss.
- Log Pyro loss as `train_elbo` and sum over steps ([1071])

Contributors

- [adamgayoso]
- [romain-lopez]
- [PierreBoyeau]
- [talashuach]
- [cataclysmus]
- [njbernstein]

Changes

- Includes new optional variance parameterization for the `Encoder` module ([1037]).
- Provides new way to select subpopulations for DE using Pandas queries ([1041]).
- Update reference to peakVI ([1046]).
- Pin Pytorch Lightning version to \<1.3

Contributors

- [adamgayoso]
- [PierreBoyeau]
- [talashuach]