pub_soft/alphafold3
3
0
Fork 0
mirror of https://github.com/google-deepmind/alphafold3.git synced 2026-06-02 11:54:36 +08:00
Code Issues Packages Projects Releases Wiki Activity
229 Commits 1 Branch 3 Tags
c20f0708597b3ca1696b65aac128ea74c43aa030
Commit Graph

229 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Augustin Zidek
c20f070859 Exit if any of the required programs is not installed 
Thanks for the suggestion, @Ashutosh0x.

PiperOrigin-RevId: 872800291
Change-Id: If1266222844de98850b7ae4141e2f50001cc4d63
 2026-02-20 02:25:13 -08:00
Augustin Zidek
703d5c8375 Make sharded Jackhmmer result merging more robust in case of e-value ties 
E-values could be equal, but bit score might have higher resolution. Use that for tie brakes before resorting to the name.

PiperOrigin-RevId: 870903887
Change-Id: I5c8648456c85ef4d1ff4dfeb15b459ba204184e3
 2026-02-16 08:30:20 -08:00
Augustin Zidek
2761bf1cfc Use the logging name instead of the one from fold_input 
PiperOrigin-RevId: 870860151
Change-Id: I1d4485e1e08c3ca790899c33190b989d103eaf56
 2026-02-16 06:13:52 -08:00
Augustin Zidek
e2b8ffd6a7 Implement relevant improvements from #615 
PiperOrigin-RevId: 868589660
Change-Id: Iac82ddf73f9f82118b935550afbe0ea13f6cd2eb
 2026-02-11 03:27:47 -08:00
Augustin Zidek
ceb32296c7 Move CONTRIBUTING.md to the root directory 
PiperOrigin-RevId: 867451876
Change-Id: Ibdaaee8902a1c1150e99225632300778c016f5f6
 2026-02-09 00:41:32 -08:00
Augustin Zidek
ebfe70a27a Update Tokamax to 0.0.10 
* Removes transitive TensorFlow dependency and others.
* Remove setuptools dependency which is no longer needed.
* Bump JAX to 0.8.2 (required by Tokamax).

PiperOrigin-RevId: 865449256
Change-Id: Iab74fa7ccf36db5c38e7c3d3d37e818a1d9d3952
 2026-02-04 09:41:35 -08:00
Augustin Zidek
c13d928a34 Update RDKit to 2025.9.4 
PiperOrigin-RevId: 865437867
Change-Id: Iaa116d661bb5743e285efcfe7115f8242f7c0db9
 2026-02-04 09:15:20 -08:00
Augustin Zidek
cf4bb5bd9e Use UV_LINK_MODE=copy to prevent Docker build issues on some platforms 
Relevant issues:
* https://github.com/google-deepmind/alphafold3/issues/603
* https://github.com/google-deepmind/alphafold3/issues/612

PiperOrigin-RevId: 865416421
Change-Id: Idaf8e531a62638bce5e0748f64a7ddc00b554677
 2026-02-04 08:18:53 -08:00
Augustin Zidek
6b50e384d3 Add the glycan modeling paper in the list of community tools 
Adapted from https://github.com/google-deepmind/alphafold3/pull/608/changes, thanks @chinchc.

PiperOrigin-RevId: 862720014
Change-Id: I021eef0b46158e8ce28e97d9172ba6e79ef3f30c
 2026-01-29 07:03:17 -08:00
Augustin Zidek
22b9ab80d3 Make AlphaFold 3 uv setup compatible with aarch64 
Addresses https://github.com/google-deepmind/alphafold3/issues/593.

PiperOrigin-RevId: 858480255
Change-Id: I7862b204300d137e367bbaaf31213f7164fdd8d1
 2026-01-20 01:56:05 -08:00
Augustin Zidek
c7a1ad5c51 Add a documentation on model parameters 
Addresses https://github.com/google-deepmind/alphafold3/issues/594.

PiperOrigin-RevId: 857083798
Change-Id: I1d8e42ab6e53075a2c05d70d10a14a8e686fa566
 2026-01-16 04:04:09 -08:00
Augustin Zidek
f6032f0f6a Add a flag to compress large output files using zstandard 
Inspired by https://github.com/google-deepmind/alphafold3/pull/544 (thank you @ntnn19), but with some differences:
1. Uses zstandard instead of gzip for faster compression times and better compression ratios.
2. Compresses only large files.
3. Doesn't compress already compressed files (distogram and embeddings).
4. Doesn't write large temporary files, compresses directly.

PiperOrigin-RevId: 856630504
Change-Id: I4413e26c9d2e8e4592c372dc675dcfa6044fc257
 2026-01-15 06:11:23 -08:00
Augustin Zidek
8d0a8db07e Make everything past model_runner mandatory kwarg in process_fold_input 
PiperOrigin-RevId: 856615431
Change-Id: I1e2a126e96e16dd2660f6e9fec9d9ffa02eeb7b2
 2026-01-15 05:15:31 -08:00
Augustin Zidek
3e2bc0ce35 Add JAAG to community tools documentation 
Addresses:
* https://github.com/google-deepmind/alphafold3/pull/583
* https://github.com/google-deepmind/alphafold3/issues/537
PiperOrigin-RevId: 856584357
Change-Id: Ifdd92878a01f30512381302cc0945cb454e9db1d
 2026-01-15 03:22:36 -08:00
Augustin Zidek
8d37fc1cb9 Modernize setup, switch to Python 3.12, and migrate to uv 
PiperOrigin-RevId: 856564478
Change-Id: I31af1c170846fb787235dfc03b3a6da0464f51be
 2026-01-15 02:14:05 -08:00
Augustin Zidek
2f40800a3a Force the legacy git backend to prevent build errors with new git versions 
PiperOrigin-RevId: 856186746
Change-Id: I447a4a3ab65674da75a908816208a6fe377232ed
 2026-01-14 06:31:52 -08:00
Augustin Zidek
d7ab550d1b Update CONTRIBUTING.md with a policy for AI-generated code 
PiperOrigin-RevId: 855264083
Change-Id: I0c6b826c2edf3a0939d03407cf6e9ddaf8d383f6
 2026-01-12 09:37:49 -08:00
Augustin Zidek
b78e215394 Update tokamax dependency to version 0.0.8 
* This fixes an issue with immutabledict dependency missing in previous versions of Tokamax.
* JAX and Haiku have to be updated as well.

PiperOrigin-RevId: 854207701
Change-Id: I8c30856504f44e43e3a058e89b0866922ca56a5b
 2026-01-09 08:39:58 -08:00
Augustin Zidek
a701815575 Upgrade GitHub Actions to the latest version (v6) 
Thanks to salmanmkc@ who suggested this in https://github.com/google-deepmind/alphafold3/pull/571.

PiperOrigin-RevId: 854113440
Change-Id: Ib71810028c7492604978eb1c2f9e46d5f57486b5
 2026-01-09 03:06:01 -08:00
Augustin Zidek
8eaae04e1d Add CONTRIBUTING.md 
PiperOrigin-RevId: 854113365
Change-Id: I1a45e75f3138d848943e1442e1d0b1787f1f699f
 2026-01-09 03:05:19 -08:00
Josh Abramson
3a09f049bf remove unused msa and template store code. 
PiperOrigin-RevId: 839187236
Change-Id: I5144812d68e0525f3ea84f32a94d0f4ab89b43ec
 2025-12-02 03:31:32 -08:00
Ryan Pachauri
389078218c Migrate AF3 to tokamax.gated_linear_unit and tokamax.dot_product_attention. 
PiperOrigin-RevId: 838262306
Change-Id: I321a78b2a7d0d5cdeabe797c59b1e9c03e33780d
 2025-11-29 19:03:26 -08:00
James Spencer
2e3703e82a fix support for sharded databases 
PiperOrigin-RevId: 836803355
Change-Id: Icc263c4a1ca153da99a71a336e1150b351cc4d27
 2025-11-25 14:01:34 -08:00
Akvile Zemgulyte
e850301268 Internal change 
PiperOrigin-RevId: 833709597
Change-Id: Ib13795c7d9edb678f07b347c44edeb1319b5510f
 2025-11-18 01:33:44 -08:00
Foromo Daniel Soromou
f3e86f27df Add safe_pickle module for secure loading of pickled data. 
This module provides functions to load pickled data with enhanced security. It restricts unpickling to a small set of allowed builtin types. This mitigates risks associated with unpickling untrusted data while maintaining performance for AlphaFold 3's generated artifacts (which is 5x faster than json).

PiperOrigin-RevId: 827973163
Change-Id: I405ac3ebc0035df8bf6bf9c66b48e5894278b98f
 2025-11-04 07:56:06 -08:00
Augustin Zidek
c0ab1c40f3 Include fully unresolved chains in Structure when going from mmCIF 
PiperOrigin-RevId: 827517473
Change-Id: I4692df12aeaf95c7507b8b4371d0474327d0824c
 2025-11-03 08:51:58 -08:00
Augustin Zidek
9ecec8d19e Handle CRLF line endings in CIF parsing 
Addresses https://github.com/google-deepmind/alphafold3/issues/546#issuecomment-3472843608

PiperOrigin-RevId: 827414127
Change-Id: I355409bf40d2869dd5fec638a7fe626677535d19
 2025-11-03 03:13:24 -08:00
Augustin Zidek
aed6e82cb1 Add an option to make entity IDs unique in structure.concat 
PiperOrigin-RevId: 826028193
Change-Id: I8af72087770fafa68135093d0c0fa55d6cbd03f2
 2025-10-30 07:46:51 -07:00
Augustin Zidek
805adc3863 Add option to run Jackhmmer/Nhmmer genetic search in sharded mode (10-30x faster) 
PiperOrigin-RevId: 825546064
Change-Id: Ib421e47bb9ca7eea512c49a532e7e995a0f5721f
 2025-10-29 07:55:10 -07:00
Augustin Zidek
46fe3f0f60 Add documentation for pre-computing and reusing MSA and templates 
PiperOrigin-RevId: 825092403
Change-Id: Idfeb86387ad3e83f5742cfa19e18f6faf3530315
 2025-10-28 10:09:19 -07:00
Augustin Zidek
9c183c8234 Make the ordering of the performance documentation more logical 
PiperOrigin-RevId: 824606495
Change-Id: Ie4e8558a95e4ddf43f2bb7486e025322f4b6860c
 2025-10-27 11:35:17 -07:00
Augustin Zidek
ec65897c60 Make unnecessarily public attributes private in tool wrappers 
PiperOrigin-RevId: 823539582
Change-Id: I846566acfd7fc089fabb1c83775f1cf55bc3bc4f
 2025-10-24 08:27:09 -07:00
Augustin Zidek
b77e80447f Increase ranking score upper bound tolerance 
PiperOrigin-RevId: 822163464
Change-Id: I93b48804b98463a2d447c2fa043bdeeef873080a
 2025-10-21 09:56:51 -07:00
Benjamin Chetioui
cf404610b5 Explicitly set backend='triton' in pallas_call invocations. 
This change is being made because the default backend for `pallas_call` is changing. Explicitly setting `backend='triton'` ensures that these calls continue to use the Triton backend, maintaining current behavior.

PiperOrigin-RevId: 822019610
Change-Id: I7ba83894fcd960d424502335c8b904d76c88a733
 2025-10-21 02:26:33 -07:00
Augustin Zidek
c9121bf646 Improve error message when testing expected scores 
PiperOrigin-RevId: 821539922
Change-Id: Iec34e6d220c5de3fcae77117ab86bbdbe871ad0e
 2025-10-20 01:28:22 -07:00
Augustin Zidek
3c89cc7b89 Improve error message and consistency of ordering 
PiperOrigin-RevId: 820694312
Change-Id: I6b04514b1f838c49bf5dca1e138e1d6b2ff81139
 2025-10-17 08:06:15 -07:00
Augustin Zidek
142e4bc6e5 Fix residue mapping bug in _maybe_add_missing_scheme_tables 
PiperOrigin-RevId: 817086740
Change-Id: Ib4a74efd410ec25f0cd92b6d6ffa23f684e660fb
 2025-10-09 02:29:51 -07:00
DeepMind
03a6c295e5 Return b factors and occupancy from to_res_arrays() 
Change to_res_arrays() to return a ResArrays dataclass which contains [num_res, num_atoms, *] shaped arrays for all atom-level data in a Structure.

PiperOrigin-RevId: 814245949
Change-Id: I38abeec96c8d4ed97a06964e6c7b7fc74f37f97d
 2025-10-02 08:31:47 -07:00
Augustin Zidek
a8ecdb2d7a Cache the underlying CCD dictionary instead of the whole CCD object. 
* Enables removal of the chemical_components.cached_ccd function.
* If there are multiple custom CCDs (tiny in comparison to the full CCD) in a single run, this will consume only O(1) RAM instead of O(num_inputs).

Addresses:
* https://github.com/google-deepmind/alphafold3/pull/514
* https://github.com/google-deepmind/alphafold3/issues/509
PiperOrigin-RevId: 804321599
Change-Id: I484f851ce79cd6b1c5c31bf70b6cd945791f8b66
 2025-09-08 01:44:10 -07:00
Augustin Zidek
85a03ec086 Use the CPU count available to the process instead of total CPU count 
Addresses https://github.com/google-deepmind/alphafold3/issues/513

PiperOrigin-RevId: 803078244
Change-Id: I6446deac260f13ad08873d1c464e65444c099283
 2025-09-04 10:20:38 -07:00
Augustin Zidek
b467f92160 Improve validation of CIF files. 
* Make sure the data name in `data_<name>` is non-empty.
* Check for duplicate key names.
* Check that the multi-line tokens are closed at the end of the file.
* Fix also two broken mmCIF files this check has uncovered.

PiperOrigin-RevId: 802572236
Change-Id: Ie7a3a5ec816ec5b97158508cc1b12064cf0e70a8
 2025-09-03 08:51:00 -07:00
Augustin Zidek
4208665547 Make sure Structure doesn't have a name that is an empty string 
PiperOrigin-RevId: 802558849
Change-Id: Ic469eec11e92d14df339d3a3d9205f144dd005fc
 2025-09-03 08:10:28 -07:00
Augustin Zidek
0ea324ce74 Validate that a non-loop CIF key has only a single value 
PiperOrigin-RevId: 802455341
Change-Id: I089c4e0a9a52862b5eba6d26023184a4a34efc96
 2025-09-03 01:48:24 -07:00
Augustin Zidek
7b816f4035 Add support for protein/DNA/RNA/ligand descriptions 
Suggested in https://github.com/google-deepmind/alphafold3/issues/496.

PiperOrigin-RevId: 802081348
Change-Id: I666466fd6a770b6f4a891ed33e6a26651d600c4a
 2025-09-02 04:11:53 -07:00
Augustin Zidek
a2b03dab51 Validate the number of tokens in a CIF loop 
PiperOrigin-RevId: 797771843
Change-Id: Ib34796c2e9f402cd2cad79d67127ddb32f0803ae
 2025-08-21 08:17:40 -07:00
Augustin Zidek
2e2ffc10ab Improve the gpu_device flag description 
Based on the suggestion in https://github.com/google-deepmind/alphafold3/issues/493

PiperOrigin-RevId: 792089424
Change-Id: Iea8d0527725c2febbc5e09b619ca7ad1f778ddba
 2025-08-07 03:53:37 -07:00
Augustin Zidek
563896c75c Improve comments in CifDict 
PiperOrigin-RevId: 791623617
Change-Id: I569b0c65d427f32bbb6a6e50a4791cb774347375
 2025-08-06 04:05:16 -07:00
Augustin Zidek
722437b4ff Fix handling of tokens that contain quotes surrounded by spaces in CifDict::ToString 
* `A "A" A` must be quoted as `'A "A" A', B 'B' B` must be quoted as `"B 'B' B"`.
* `C "C" 'C' C` must be quoted as `\n;C "C" 'C' C\n;`.
* Previously, `CifDict` didn't quote these correctly and e.g. `A "A" A` would become three tokens because of the bad early return on the first space in the token.

This makes `CifDict::ToString` ~15% faster as the trivial check allows us to skip the extra newline check for most tokens.

Reported in https://github.com/google-deepmind/alphafold3/issues/471.

PiperOrigin-RevId: 791587762
Change-Id: If6616263e570dff534633bd0302afd23471f946f
 2025-08-06 02:05:23 -07:00
Augustin Zidek
f2edd59f87 Undo backwards incompatible Triton API change 
PiperOrigin-RevId: 790639906
Change-Id: If98343e8a5601baeeb882229a4e4c5dcdc601881
 2025-08-04 01:37:39 -07:00
Augustin Zidek
903a931d36 Improve documentation on running data pipeline and inference separately 
Prompted by https://github.com/google-deepmind/alphafold3/issues/485.

PiperOrigin-RevId: 789700619
Change-Id: I0e876a1130cc537421e8916fcad1230302196b64
 2025-08-01 03:56:58 -07:00