Alphafold Versions Save

Change log

• More robust download in Colab with shutil (thanks @gmihaila).
• Added ability to only run relax for the best unrelaxed model in the run_alphafold.py.
• Improved documentation for the ranked outputs (thanks @ulupo).
• Removed jax dependency from results pkl.
• Updated tensorflow to 2.11.0.
• Improved documentation on how to install aria2c (thanks @janxkoci).
• Made _chem_comp.type logic case-insensitive for mmCIF parsing.
• Improved error messages when cells are submitted out of order in Colab (e.g. when the runtime restarts).
• Fixed incorrect type annotations.
• Bumped Python to 3.9 in Colab.
• Improved robustness of masked softmax for bfloat16.
• Bumped pyopenssl in Colab to patch cryptography dependency issue.

Version v2.3.1 combines a few small updates.

Change log

• Add option for eval time dropout.
• Add A100 inference timings to README.
• Speed up MSA lookups in Colab for multimers.
• Increase max allowed sequence length in Colab to 4,000.
• Improve documentation on initial install and run of AlphaFold.
• Documentation improvements and other small fixes (thanks @eltociear).
• Pin part of violations computation on CPU. Fixes GPU memory issues during relax stage.

Version v2.3.0 updates the AlphaFold-Multimer model parameters. These new models are expected to be more accurate on large protein complexes but use the same model architecture and training methodology as our previously released AlphaFold-Multimer paper. See the v2.3.0 release note for more details.

Thanks to various memory optimisations, AlphaFold-Multimer now uses less GPU memory and it can therefore handle longer proteins.

A number of other bug fixes and small improvements have been made.

Change log

• Added new AlphaFold-Multimer models with better accuracy on large protein complexes.
• Added early stopping to recycling.
• Added filtering for non-protein sequences in the pdb_seqres download script to prevent template search errors.
• Fixed a bug where histidine residues had sometimes swapped atom coordinates after relaxation (thanks @avwillems).
• Updated MGnify to 2022_05, UniRef90 to 2022_01, UniClust30 to 2021_03, UniProt in Colab notebook to 2021_04.
• Used bf16 in multimer inference – reduces GPU memory usage.
• Upcast to fp32 when using bf16 in LayerNorm and replace hk.LayerNorm with common_modules.LayerNorm.
• Updated Jax to 0.3.25 and Haiku to 0.0.9 for consistency with the AlphaFold Colab notebook.
• Changed TriangleMultiplication to use fused projections and various other memory optimisations.
• Upgraded Python version in the AlphaFold Colab notebook to 3.8.
• AlphaFold Colab notebook usability improvements – multimers with up to 20 chains are now supported, higher sequence length limits, number of recycling iterations can now be controlled, and added an option to run single chains on the multimer model.
• Relaxation metrics are now saved in relax_metrics.json.
• Some Jax deprecation errors were addressed (thanks @jinmingyi1998).
• Various documentation and code improvements (thanks @mathe42).

Version v2.2.4 is a bug fix release

Change log

• Bump versions of third party libraries: jax 0.3.17, absl-py 1.0.0, haiku 0.0.7, numpy 1.21.6, tensorflow 2.9.0
• Adapt jnp.take to account for behaviour with the new jax version, see https://github.com/deepmind/alphafold/issues/513 (thanks @sokrypton).
• Reduce size of docker image by removing package caches, see https://github.com/deepmind/alphafold/pull/526 (thanks @TheDen).
• Fix incorrect argument in backbone_loss, see https://github.com/deepmind/alphafold/issues/570 (thanks @sokrypton).

Version v2.2.3 is a bug fix release.

Change log

• Pin Conda version to 4.13.0 to prevent Docker/Colab setup issues (thanks @Meghpal, @michaelkeith18).
• Change the Colab PAE json output to new format that matches the format used in the new release of the AlphaFold Protein Structure Database (AFDB). See the AFDB FAQ for a description of the new format.
• Add a readme file for AFDB.
• Type hint improvements.
• Fix tests and improve internal testing infrastructure.
• Fix Dockerfile breakage due to https://github.com/google/jax/issues/11142.

A small bug fix release that fixes a bug introduced in v2.2.1 (thanks @lucajovine).

Version v2.2.1 is a bug fix release.

Change log

• Update from CUDA 11.1 to to 11.1.1 which addresses the public key issue.
• Pin protobuf version to 3.20.1 (thanks @britnyblu, @ShoufaChen, @aputron).
• Clarify in the README that AlphaFold works only under Linux.
• Fix the jax.tree_multimap deprecation warning.
• Do not reuse the temporary output directory in run_alphafold_test (thanks @branfosj).
• Fix the version in setup.py (thanks @cmeesters).

Version v2.2.0 updates the AlphaFold-Multimer model parameters. These new models have greatly reduced numbers of clashes on average and are slightly more accurate. Read the updated AlphaFold-Multimer paper for more details.

A number of other bug fixes and small improvements have been made.

Change log

• Added new AlphaFold-Multimer models with greatly reduced numbers of clashes on average and slightly increased accuracy.
• Use DeviceRequest rather than runtime=nvidia to expose GPUs to the container (thanks @aburger).
• Simplified mounting of files in Docker.
• Removed unused bias argument in GlobalAttention (thanks @breadbread1984).
• Removed prokaryotic MSA pairing algorithm as it didn’t improve accuracy on average.
• Added the ability to run with multiple seeds per model to match the AlphaFold-Multimer paper.
• Fixed degraded performance when using num_recycle=0 with models trained with recycling due to incorrect skipping of layers (thanks @sokrypton).
• Added split_rng=False (current default) to sharded_map to support new Haiku release.
• Removed unused code in amber_minimize.py.

Version v2.1.2 is a bug fix release that also includes the earlier license change.

Change log

• Update the license of the AlphaFold parameters from CC BY-NC 4.0 to CC BY 4.0. There are no changes to the actual model parameters.
• The relaxation stage now runs on GPU by default and should be roughly 3x faster thanks to that. You can control this behaviour using the enable_gpu_relax flag (thanks @ojcharles).
• The relaxation stage can now be disabled using the run_relax flag (thanks @bkpoon).
• AlphaFold in Docker is now run as the current user not as the root, you can control that using the docker_user flag (thanks @akors).
• Truncate the MSA when reading the raw Stockholm file to prevent out of memory issues. This should help in cases where the MSA found by Jackhmmer was massive (thanks @hegelab).
• Update Dockerfile CUDA version to 11.1 and fix JAX version (thanks @chrisroat).
• Small README, Colab, and flag documentation improvements.

Version v2.1.1 is a bug fix release for the AlphaFold-Multimer release (v2.1.0).

Change log:

• Fixed a bug which caused a crash if the multimer input fasta file contained SwissProt identifiers in the sequence descriptions (thanks @arashnh11, @RodenLuo).
• Fixed a bug in the Colab notebook with single-chain PAE visualisation (thanks @Alleko).
• A few README clarifications and additions.