New features added:
1. Refactorised the codes; introduced folding_backend 303 thanks for the help from maurerv and DimaMolod
2. Allow the user to pad input matrices to desired number of MSAs and desired number of residues to speed up overall modelling process and avoid unnecessary re-compiling of AlphaFold neural network.
3. New way of modelling with customised structure templates without the need of recalculating the features again. 268
4. Separated post-modelling processes from prediction process 297 by DimaMolod
5. Supports full mmseqs2 mode i.e. without the need of local structural template database when using mmseqs mode 233
Bugs fixed:
1. Fixed incorrect colour scheme when ploting structures in jupyter-notebook 304 thanks for the help from gchojnowski and report from gilep
2. Fixed operands broadcast error when the features are created by mmseqs2 287 thanks for the report from Qrouger
3. Updated alpha-analysis.sif to avoid crashes when no model satisfies the cutoff value 307
4. Fix the config.cfg to avoid installing tensorflow versions that are not compatible with GPUs without latest CUDA. 298 Thanks for the help from kashyapchhatbar
Notice:
Apart from installing the beta version of alphapulldown from pypi, using pip install alphapulldown==2.0.0b1, please re-download the alpha-analysis singularity images again.
If your results are from AlphaPulldown prior to version 1.0.0, please use the link: [`alpha-analysis_jax_0.3.sif`](https://www.embl-hamburg.de/AlphaPulldown/downloads/alpha-analysis_jax_0.3.sif).
If your results are from AlphaPulldown with version >=1.0.0, please use the link: [`alpha-analysis_jax_0.4.sif`](https://www.embl-hamburg.de/AlphaPulldown/downloads/alpha-analysis_jax_0.4.sif).