Release v0.1.4 - compatibility with scvi-tools 1.0.0

option to use torch.compile (#336)

v0.1.3 release

This release fixes a numerical stability issue by reverting back to using the prior mean (rather than the median of 500 samples from the prior distribution) to initialise posterior loc.

v0.1.2 release

Main changes:

1. SpaceJam model for Nanostring data (cell2location extended to additional technical effects) reimplemented in Pyro by @AlexanderAivazidis.
2. Option for memory-efficient posterior distribution quantification.
3. Extended architecture of amortising NN.
4. Compatibility with latest scvi-tools.
5. Minor bug fixes.

v0.1 release

This release introduces changes listed below:

Big changes to code:

• Updates for compatibility with scvi-tools>=0.16.0
• Deleting pymc3 code
• Revised tutorial (including per cell type expression prediction
• Amortised NN using AutoGuideMessenger and option to use hierarchical guide

Changes to defaults:

• Changing our default detection_alpha because within-slide tech effects seem to be more prevalent than their absence (detection_alpha=20) and recommend to run both detection_alpha=20 and detection_alpha=200 in the tutorial.

Changes to docs

• Template issue: bug (version, are you following scvi-tools tutorial)
• Template issue: usage question (version, are you following scvi-tools tutorial, which technology was used to generate reference data / spatial data, how many cell types, how many batches, how many genes, how many spatial locations)
• Update tutorial to match scvi-tools tutorial
• Update tutorial according to this feedback #65 (comment)
• Delete pymc3 tutorials from documentation website (reorder the rest to make it easier to understand that pymc3 should not be used)
• Fix documentation website for various classes including user-facing Cell2location class (automodule->autoclass)
• Update readme - shorten what's possible
• Update readme - future work
• Update readme - add link to Nat Biotech paper
• Update acknowledgements
• Update copyright date
• Revise colab version of the notebook (less posterior samples)
• Add this line to tutorial: "The values are stored in adata.uns[f"mod_coloc_n_fact{n_fact}"] in a similar output format main cell2location results."

Pre-release: normalisation model, plot improvements, bug fixes

This is the last pre-release before we introduce cell2location based on pyro and scvi-tools. The main purpose of this release is fixing package versions and dependencies for pymc3 implementation of models - which helps to provide docker image that reproducibly works with older models and older user interface (run_cell2location / run_regression / etc).

This pre-release includes 1) the model which can normalise within experiment variation in RNA detection sensitivity, 2) improvements to data visualisation, 3) major refactoring of how models are organised, 4) bug fixes (no critical bugs).

Pre-release with updated tutorials, performance improvements

This pre-release includes updated tutorial notebooks, new Google Colab notebook, performance improvements (run_c2l and plot_spatial).

Pre-release with gene selection and minor improvements

This pre-release adds gene selection by specificity to cell types, documentation and notebook improvements, some helpful error messages.

Pre-release with pyro translations and refactored models

This pre-release contains pyro translation of cell2location, improved documentation and major refactoring of variable names incompatible with earlier versions.

Pre-release with multi-sample models

Big update to cell2location documentation and renaming of model classes

• added tutorial about estimating reference signatures of cell types
• renamed model classes for readability
• pipelines can automatically select models (e.g. model with technical effects & multi-sample).