biolists / folding_tools

A collection of *fold* tools

📖 Table of contents

• Predictors
• Tools and Extensions
• Databases and Datasets
• Webservers
• Discontinued

💡 Notes

• The following lists are curated by humans, as such may be incomplete
• We only include software targeting the folding problem combining learnings from AlphaFold 2 and protein language models. You may find other ML on protein tools at Kevin's incredible ML for proteins list.
• We do not wish to advertize one tool over any other, but simply list the tools we are aware of in either random or alphabetical order
• Any suggestions for improvements and additions are welcome as issues or pull requests
• Projects we identify as discontinued are marked with 💀 and in a section at the end

⚡️ Brought to you by:

• @sacdallago
• @sokrypton

---

Predictors

[in alphabetical order]

• MSA-based (uses Multiple Sequence Alignments (MSAs) as input)

  • AlphaFold2
    • The original AlphaFold 2 method
    • Features: monomer, multimer
    • Other: Colab Notebook
  • ColabFold
    • Faster AF2 compiling and MSA generations
    • Features: monomer, multimer
    • Other: localcolabfold
  • FastFold
    • Runtime improvements to OpenFold (see below)
    • Features: monomer
  • HelixFold
    • Reimplementation of AF2 in PaddlePaddle
    • Features: monomer
  • MEGA-Fold
    • Reimplementation of AF2 in MindSpore; provides training code, training dataset and new model params.
    • Features: monomer
  • OpenFold
    • Reimplementation of AF2 in PyTorch; provides training code, training dataset and new model params.
    • Features: monomer
    • Other: Colab Notebook
  • RoseTTAFold
    • Reproduced AF2 in PyTorch before details of AF2 were available; new model parameters.
    • Features: monomer
    • Other: Unofficial Colab Notebook
  • Uni-Fold
    • Reimplementation of AF2 in PyTorch; provides training code and new (monomer/multimer) model parameters.
    • Features: monomer, multimer
    • Resources: Colab Notebook
  • Uni-Fold-jax
    • Implementation of AF2's training code.

• pLM-based (using embeddings from protein Language Models (pLMs) as input)

  • ESM-Fold
    • Features: monomer
    • Other: Unofficial Colab notebook,API
  • EMBER3D
    • Features: monomer
  • HelixFold-single
    • Features: monomer
    • Resource: https://paddlehelix.baidu.com/app/drug/protein-single/forecast
  • IgFold
    • pLM focused on antibody sequences
    • Features: monomer
    • Other: Colab Notebook
  • OmegaFold
    • Features: monomer
    • Other: Unofficial Colab Notebook, [tweet] Martin comparing structures, [tweet] Sergey's positional encoding observation
  • RGN2 Github
    • Uses AlphaFold's structure module (AF2Rank's template protocol) for final structure refinement.

• Other

  • EquiFold
    • Diffusion model to predict protein structures (specifically antibodies)
    • Features: monomer

Tools and Extensions

• gget (AF2)
• alphafold_finetune
  • finetune AlphaFold for Protein-Peptide prediction
  • Other: [tweet] Amir's announcement
• AlphaPulldown
  • protein-protein interaction screens using AlphaFold-Multimer
• ColabDesign
  • Backprop through AlphaFold for protein design
• AF2Rank
  • Rank Decoy Structures/Sequences using AlphaFold
  • Resource: Colab Notebook
• protein_structure_module_of_AF2
  • IPA implementation in pytorch

---

Databases of predictions

• AlphaFold Database
  • All sequences in UniRef90 - viral sequences; Based on AlphaFold 2
  • Resource: https://alphafold.ebi.ac.uk
• Eukaryotic interactormes
  • Protein-Protein interactions; Based on RoseTTAFold and AlphaFold 2
  • Resource: https://www.ebi.ac.uk/pdbe/news/predicted-complexes-modelarchive-now-pdbe-kb-pages
• Structures of human-transcriptome isoforms
  • Based on ColabFold (AlphaFold 2)
  • Resource: https://www.isoform.io
• AlphaFill
  • Enriching the AlphaFold models with ligands and co-factors (AlphaFold 2)
  • Resource: https://alphafill.eu/
• IgFold Database
  • Predictions specific to antibody sequences; based on OAS dataset and IgFold
  • Resource: https://data.graylab.jhu.edu/igfold_oas_paired95.tar.gz

Datasets for training

• OpenProteinSet
  • 16 million MSAs for PDB, and AF2 structure predictions
  • Manuscript: https://arxiv.org/abs/2308.05326
  • Resource: https://registry.opendata.aws/openfold/
• MindSpore
  • MSAs for 570K PDBs + 745K Distillation
  • Manuscript: https://arxiv.org/abs/2206.12240
  • Resource: http://ftp.cbi.pku.edu.cn/psp/

---

Webservers

• Lambda PredictProtein
  • Based on ColabFold; Limited to sequences up to 500AAs
  • Resource: http://embed.predictprotein.org/
• Robetta
  • Based on RoseTTAFold
  • Resource: https://robetta.bakerlab.org/

---

Discontinued

• 💀 Moonbear
  • Resource: https://www.getmoonbear.com/
  • Other: [tweet] Stephanie's announcement
• 💀 Lucidrains' AlphaFold2
  • AF2 reproduction attempt
  • Features: monomer
• 💀 Lupoglaz's OpenFold2
  • AF2 reproduction attempt
  • Features: monomer

folding_tools's Issues

Add a "Features" bullet-point

Hi all,

Thanks for starting this repo, I think it's a great way to have everything going on in one place.

Could I suggest that each entry has a "Feature" bullet-point, which includes key:value pairs like:

Method XYZ
- Paper: URL
- Code repo: URL
- Features = Model_input:MSA; Multimers:no
Method ABC
- Paper: URL
- Code repo: URL
- Features = Model_input:pLM; Multimers:yes

Something like that. My idea is to contribute to the repo but also have a table format based on features, a bit like the ColabFold tables, e.g. see below:

See a manually-typed table I created in Google Sheets:

Thanks for considering

Add OpenFold database

OpenFold author here---I thought you might want to include our database of folding training data + self-distillation predictions: https://registry.opendata.aws/openfold/

TmAlphaFold database

We created a database, called Transmembrane AlphaFold database (TmAlphaFold database), where membrane localization was calculated and evaluation of AlphaFold2 predicted alpha-helical transmembrane protein structures was given. We applied TMDET, a simple geometry-based method to visualize the likeliest position of the membrane plane. In addition, we calculated several parameters to evaluate the location of the protein into the membrane. This also allows TmAlphaFold database to show whether the predicted 3D structure is realistic or not. The TmAlphaFold database is available at https://tmalphafold.ttk.hu/ and has been published in NAR (https://pubmed.ncbi.nlm.nih.gov/36318239/)
Please consider to include TmAlphaFold database in your database list.

recommend a reproduction repo of alphafold2 structure module

Hi. I personally reproduced the structure module of alphafold2, which basically follows the openfold. I simplify the required pkgs and made it runable with pytorch. The code is for educational purpose. Are you interested in including it in this repo?

https://github.com/pengzhangzhi/protein_structure_module_of_AF2

Submission of folds and tangentials

I could not spot a way to submit (=shameless advertise one's own) repos, so I am opening an issue for this —please close it if it is not welcome!

Long term strategy

Since this repo is now addressing multiple topics (folding, design, pLMs), it may make sense to:

1. Have a general README that redirects to the various sub-lists
2. Transfer this repo to an agnostic organization (for this purpose I prospectively created https://github.com/biolists) so that it's not "sacdallago" who owns it -- but the community
3. Fork into that org other lists, past or future, e.g. @yangkky 's Machine learning for proteins: https://github.com/yangkky/Machine-learning-for-proteins

CC @sokrypton @duerrsimon @noeliaferruz