AF-Pipeline

---

Installation

• Clone the repository

  git clone --recursive https://github.com/isblab/af_pipeline.git

  If cloned without --recursive flag, do the following.

  git submodule init
  git submodule update

• Run setup.py

  python setup.py

---

Overview

AF-Pipeline is a package to assist in AlphaFold2¹ and AlphaFold3² related tasks. These include:

• Creating input files for prediction (for AlphaFold server³ or AlphaFold2¹ or Colabfold⁴)
• Ranking the predictions based on confidence metrics
• Extracting confidently predicted regions from the predictions (based on pae_to_domains⁵ package by Tristan Croll)
• Extracting interacting regions from the predictions

The workflow for the entire pipeline can be viewed here.

See also:

• Changelog
• Contributing
• Documentation

---

Usage

• Refer to the scripts available in the examples directory. Before running the scripts, make sure you unzip AF_predictions.zip

  unzip ./input/AF_predictions.zip -d input/

• Create JSON job files for AlphaFold server ³.

  python create_af_jobs.py \\
      -i ./input/config.yaml \\
      -o ./output/af_input_jobs \\
      -p ./input/protein_sequences.fasta \\
      -n ./input/nucleic_acid_sequences.fasta \\
      -t AF3 # you can replace this with AF2 or ColbFold

• Upload the JSON files to AlphaFold server ³ and download and extract the results.

• Rank the output predictions.

  python rank_af_predictions.py \\
      -i ./output/af_input_jobs.json \\
      -o ./output \\
      -t AF3 \\
      --pred_dirs ./input/AF_predictions/AF3

• Extract interacting patches from the predictions.

  python extract_interacting_patches.py \\
      --i ./input/best_af_predictions.json \\
      --o ./output/interacting_patches \\
      --interaction_pae_cutoff 5.0 \\
      --plddt_cutoff 70.0 \\
      --contact_threshold 8.0

• Extract confidently predicted regions from the predictions.

  python extract_rigid_bodies.py \\
      -i ./input/best_af_predictions.json \\
      -o ./output/rigid_bodies \\
      --apply_plddt_filter True

---

Workflow

graph TD
    A([Create job files for AlphaFold Server]) e1@--> B[[AlphaFoldServer]]
    click B "af_pipeline/af_input/alphafold3.html#AlphaFoldServer" "AlphaFoldServer" _blank
    B e2@--> C[/Input FASTA or JSON files for AlphaFold/] e3@--> D[\Submit jobs to AlphaFold server/]
    click D "https://alphafoldserver.com" "alphafoldserver" _blank
    D e4@--> E[/Output files from AlphaFold server/]
    E e16@--> R[[RankAF3JobSet]]
    click R "af_pipeline/rank_predictions/rank_af.html#RankAF3JobSet" "RankAF3JobSet" _blank
    R e17@--> S[/best prediction/]
    S e5@--> F[[Initialize]]
    click F "af_pipeline/parser/initialize.html#Initialize" "Initialize" _blank
    F e6@--> G[[RigidBodies]]
    click G "af_pipeline/rigid_bodies/rigid_bodies.html#RigidBodies" "RigidBodies" _blank
    F e7@--> H[[Interaction]]
    click H "af_pipeline/interaction/interaction.html#Interaction" "Interaction" _blank
    G e8@--> I[extract_rigid_bodies]
    click I "af_pipeline/rigid_bodies/rigid_bodies.html#RigidBodies.extract_rigid_bodies" "extract_rigid_bodies" _blank
    N e9@--> J[save_rigid_bodies]
    click J "af_pipeline/rigid_bodies/rigid_bodies.html#RigidBodies.save_rigid_bodies" "save_rigid_bodies" _blank
    N e10@--> K[assess_rigid_bodies]
    click K "af_pipeline/rigid_bodies/rigid_bodies.html#RigidBodies.assess_rigid_bodies" "assess_rigid_bodies" _blank
    H e11@--> M[save_ppair_interaction]
    click M "af_pipeline/interaction/interaction.html#Interaction.save_ppair_interaction" "save_ppair_interaction" _blank
    I e12@--> N(["Confidently predicted regions (rigid bodies) from AlphaFold predictions"])
    J e13@--> O[/"Output files for rigid bodies (txt/json and pdb)"/]
    K e14@--> P[/"Output files for rigid body assessment (xlsx)"/]
    M e15@--> Q[/"Output files for interacting patches (xlsx and plots)"/]
    classDef animate stroke-dasharray: 9,5,stroke-dashoffset: 900,animation: dash 25s linear infinite;
    class e1,e2,e3,e4,e5,e6,e7,e8,e9,e10,e11,e12,e13,e14,e15 animate

Loading

---

Additional Information

License: GPLv3

Testable: Yes

Footnotes

1. Jumper, J. et al. Highly Accurate Protein Structure Prediction with Alphafold. Nature 596, 583–589 (2021). ↩ ↩²

2. Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493–500 (2024). (https://alphafoldserver.com/) ↩

3. AlphaFold Server. alphafoldserver.com. Available at: https://alphafoldserver.com/. ↩ ↩² ↩³

4. Mirdita, M. et al. ColabFold: making protein folding accessible to all. Nature Methods 19, 679–682 (2022). (https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/AlphaFold2.ipynb) ↩

5. pae_to_domains: Graph-based community clustering approach to extract protein domains from a predicted aligned error matrix. Tristan Croll (https://github.com/tristanic/pae_to_domains) ↩