v0.8.7

• Fix protein plddt bfactor
• Unify build with hatch-conda-build

Full Changelog: v0.8.6...v0.8.7

v0.8.6

• Fix boltz confidence retrieval
• Properly handle when model metadata input tokens is null, e.g. msa for AF2
• Some documentation fixes
• Add metadata for fold

v0.8.5

• Fixed boltz affinity
• Fixed predictor train with poet prompt
• Add paging to listing assays and jobs
• bfactor for plddt
• Some typing improvements

v0.8.4

• Fixed broken import

v0.8.3

• Release Rfdiffusion interface
• Fixes refactor of fitting umap/svd/predictor
• Lots of docstring updates

NOTE v0.8.2 release got botched so we are skipping a number. please do not use if you find it

v0.8.1

• make single sequence mode be used on a per-protein bases
  • use protein.msa = Protein.single.sequence.mode

v0.8.0

• refactored into domain based layout
  • rewrote unit tests
• added more docstrings and audited a lot of it
• removed all deprecated workflows that used the old train and predict
• added boltz support for structure prediction
• added decoder type parameter on poet2

v0.7.0

• Added PoET 2 extra functionality around usage of query with the prompt. Refer to the example on inverse folding.
• Introduction of a Protein class for more consistent manipulation of protein objects.
• Deprecated old PoET interfaces in favour of unified access through the embedding module.
  • Usage throws an exception with info on where to access the function

v0.6.1

• Fixed issue with predictor train due to traingraph being null when starting train
• Added shortcuts for DesignConstraints

v0.6.0

• release session.predictor module
  • allows training gaussian process models on different types of embeddings
  • can also use fitted svds for high fidelity custom sized embeddings to fit GPs
  • predictors can then be used for design
  • deprecated session.train into session.deprecated.train
• release newer session.design interface
  • allows using the newly trained predictor
  • supports syntactic sugar for quickly creating criteria
  • deprecated old session.design interface which requires awkward construction using pydantic into session.deprecated.design
    • might still be needed to support older train models
• migration is planned for all old models and design into the new interface so that they can be listed, fetched and used for prediction.