ColabFold

The setup-and-start-local.sh script automates the setup of a local ColabFold API server on Linux (or macOS for testing). It performs the following tasks:

• Verifies required software (curl, aria2c, rsync, aws).
• Downloads platform-specific MMseqs2 and mmseqs-server binaries.
• Downloads databases (UniRef30 and ColabFoldDB).
• Downloads and compiles the API server binary.
• Starts the API server.

Important Configuration Steps before running:

1. Enable GPU (Linux only): Uncomment the export GPU=1 line inside the script. This adds --paths.colabfold.gpu.gpu 1 --paths.colabfold.gpu.server 1 to the server startup.
2. Select PDB Mirror: At the top of the script, uncomment exactly one pair of PDB rsync mirrors (RCSB, PDBe, or PDBj). The script will exit if no mirror is selected.

Quick Test Mode: To start the server with templates disabled and a tiny database for rapid testing, set the DEBUG_MINI_DB=1 environment variable.

The following bash script installs the necessary Python packages, clones the ColabFold script, sets up the AlphaFold repository, and downloads the model parameters.

Note: The script includes a patch to remove END and ENDMDL from PDB lines in alphafold/common/protein.py to prevent Biopython compatibility errors.

# Example of the dependency installation logic used in the notebook
pip -q install biopython dm-haiku ml-collections py3Dmol

wget -qnc https://raw.githubusercontent.com/sokrypton/ColabFold/main/beta/colabfold.py

if [ ! -d "alphafold/" ]; then
  git clone https://github.com/deepmind/alphafold.git --quiet
  mv alphafold alphafold_
  mv alphafold_/alphafold .
  sed -i "s/pdb_lines.append('END')//" /content/alphafold/common/protein.py
  sed -i "s/pdb_lines.append('ENDMDL')//" /content/alphafold/common/protein.py
fi

if [ ! -d "params/" ]; then
  wget -qnc https://storage.googleapis.com/alphafold/alphafold_params_2021-07-14.tar
  mkdir params
  tar -xf alphafold_params_2021-07-14.tar -C params/
  rm alphafold_params_2021-07-14.tar
fi

To use the experimental OmegaFold hacks, you must install the OmegaFold repository, its requirements, py3Dmol, aria2, hhsuite, and the colabfold.py script. This setup also involves downloading the model checkpoint release1.pt to ~/.cache/omegafold_ckpt/model.pt.

Note: This setup is designed for a Colab-like environment where apt-get and pip are available.

import os,sys,re
from IPython.utils import io
if "SETUP_DONE" not in dir():
  import torch
  device = "cuda" if torch.cuda.is_available() else "cpu"
  with io.capture_output() as captured:
    if not os.path.isdir("OmegaFold"):
      %shell git clone --quiet https://github.com/sokrypton/OmegaFold.git
      %shell cd OmegaFold; pip -q install -r requirements.txt
      %shell pip -q install py3Dmol
      %shell apt-get install aria2 -qq > /dev/null
      %shell aria2c -q -x 16 https://helixon.s3.amazonaws.com/release1.pt
      %shell mkdir -p ~/.cache/omegafold_ckpt
      %shell mv release1.pt ~/.cache/omegafold_ckpt/model.pt
      %shell wget -qnc https://github.com/soedinglab/hh-suite/releases/download/v3.3.0/hhsuite-3.3.0-SSE2-Linux.tar.gz
      %shell tar xfz hhsuite-3.3.0-SSE2-Linux.tar.gz
      %shell wget -qnc https://raw.githubusercontent.com/sokrypton/ColabFold/main/colabfold/colabfold.py
  os.environ['PATH'] += ":/content/bin:/content/scripts"
  SETUP_DONE = True

To use the advanced ESMFold notebook, you must install the required libraries and download the model parameters. This process installs esmfold, openfold, and several dependencies like omegaconf, pytorch_lightning, and biopython. It also downloads the esmfold.model file using aria2c.

Note on Colab Limitations: On a Tesla T4 GPU (standard free Colab), the maximum total sequence length is approximately 900 residues.

import os, time
import torch
if not os.path.isfile("esmfold.model"):
  os.system("apt-get install aria2 -qq")
  os.system("aria2c -q -x 16 https://colabfold.steineggerlab.workers.dev/esm/esmfold.model &")

  print("installing libs...")
  os.system("pip install -q omegaconf pytorch_lightning biopython ml_collections einops py3Dmol modelcif")
  os.system("pip install -q git+https://github.com/NVIDIA/dllogger.git")

print("installing openfold...")
  os.system(f"pip install -q git+https://github.com/sokrypton/openfold.git")

print("installing esmfold...")
  os.system(f"pip install -q git+https://github.com/sokrypton/esm.git@beta")

  if not os.path.isfile("esmfold.model"):
    os.system("aria2c -q -x 16 https://files.ipd.uw.edu/pub/esmfold/esmfold.model")
  else:
    while os.path.isfile("esmfold.model.aria2"):
      time.sleep(5)

if "model" not in dir():
  model_path = "esmfold.model"
  model = torch.load(model_path, weights_only=False)
  model.cuda().requires_grad_(False)