BERTose and BERToseBind workbench
The initial-submission notebook files remain preserved.
Open BERTose_workbench.ipynb, download it and run it in Jupyter, or upload the downloaded notebook into Google Colab. The notebook installs the shared code from supanthadey1/bertose-affinose-training-code and offers sequence embedding, fitted binding prediction and supported IAR completion workflows. Its checks and input limits are described in the notebook; it does not retrain the models.
For a local environment, use Python 3.12 and install requirements.txt, then open the notebook. Its code-download cells target the main revision. Scientific weights and data retain their pinned hashes and are downloaded only when requested. The binding cached example needs the released protein HDF5 (about 2.33 GB); the IAR full checkpoint needs several GB of CPU memory.
See the shared code README and training/reproduction documentation. Original code and revised model weights use Apache-2.0; author-generated data use CC-BY-4.0, while inherited inputs and dependencies retain their own terms. No complete training or benchmark replay is claimed.