• Adding improved proxy configuration for python client

• Adding support for UdonPred model (python, dart)
• Adding specialized test for half precision mode and the UdonPred predictor (python)
• Improving active learning API integration and enum handling (dart)

Key Changes:

1. Python:

   • Added merge functionality to EmbeddingsResult for simpler merging of batched results.
   • Enabled FASTA file input support for the embed endpoint.
   • Refactored batching functionality and updated examples (amylase and exotox_reproduction).
   • Refactored embed endpoint to enforce sequence limits.
   • Updated tests to cover more models, excluding VespaG due to model size constraints.
   • Adapted API following removal of plm_eval module.

2. Dart:

   • Added a new research and service stats API.
   • Improved support for research and service stats API, including updates for biotrainer features.
   • Refactored API to align with updates related to plm_eval module removal.

Bug Fixes:

• Adjusted Python API tests to align with data models and metrics.
• Resolved issues with al_simulation API test and local_only_default handling in tests.

• Fixes:

  • Added missing serializers for the prediction task in the Dart API.
  • Adapted to the refactored biocentral_service API name.
  • Preferred local URLs over remote URLs in the Python client.

• Features:

  • Added warnings about sequence lengths if they exceed the recommended limit.
  • Introduced versioning checks to the Python API for remote and local servers.
  • Added a retry mechanism for handling rate-limiting scenarios.
  • Added an active learning simulation API.
  • Added automatic git add to the generation script after client generation.
  • Implemented pruning of outdated code files after generation.
  • Added basic test of python API

• Refactors:

  • Improved the Dart and Python APIs to align with the new protocol enums.
  • Removed unnecessary protocol endpoints in both Dart and Python APIs.
  • Made updates to the SequenceTrainingData and simulation result models, improving model tracking and embedding progress.
  • Transitioned to a common embedder naming system with the CommonEmbedder enum.
  • Updated the usage of the active learning module and changed from bay_opt to the updated module.
  • Enhanced the handling of iteration results using iteration numbers for better tracking.

Initial API release

Contains API packages for python and dart.

Main Features

• Embedding: Embed protein sequences into a vector space using a protein language model.
• Predict: Predict protein properties using pre-defined prediction models.
• Train: Train your own custom prediction models (using biotrainer)
• Inference: Run inference on new data using your own custom prediction models.