AlphaFold ColabFold

2.3.13 verified Fri May 01 auth: no python

An implementation of the inference pipeline of AlphaFold v2.3.1, with patches for ColabFold. This package allows running AlphaFold2 predictions (as published in Nature) with ColabFold enhancements. Current version 2.3.13. Release cadence is irregular, tied to AlphaFold updates.

pip install alphafold-colabfold

Common errors

error ModuleNotFoundError: No module named 'alphafold' ↓

cause Misspelled or wrong package installed; package is 'alphafold-colabfold' not 'alphafold'.

fix

pip install alphafold-colabfold

error ValueError: No such model: 'model_1' ↓

cause Model name must match one of the available model filenames (e.g., 'model_1' is for CASP14, but actual parameter file might be named differently).

fix

Use one of the following model names: 'model_1', 'model_2', 'model_3', 'model_4', 'model_5' or 'model_1_ptm', 'model_2_ptm', etc. Check the parameter directory.

Warnings

breaking The package name 'alphafold-colabfold' may be confused with the older 'alphafold' package. Ensure you install 'alphafold-colabfold' specifically to get ColabFold patches. ↓

fix pip uninstall alphafold && pip install alphafold-colabfold

gotcha The run_alphafold function requires pre-downloaded AlphaFold databases (e.g., BFD, MGnify, PDB70, PDB, Uniref30, Uniclust30, etc.) and model parameters. The package does not download them automatically. ↓

fix Follow the AlphaFold setup guide at https://github.com/google-deepmind/alphafold to download databases and parameters, then set ALPHAFOLD_DB_DIR and ALPHAFOLD_MODEL_DIR environment variables.

deprecated The use_amber flag may be deprecated in future versions. The ColabFold patches recommend using 'use_amber=True' for better accuracy, but check notes on potential deprecation. ↓

fix Keep using use_amber=True unless future release notes indicate removal.

Imports

AlphaFold2
wrong
```
from alphafold import AlphaFold
```
correct
```
from alphafold.model import model as alphafold_model
```
AlphaFold is not a class in this package; model is accessed via alphafold.model.model

run_alphafold

wrong

from alphafold import run_alphafold

correct

from alphafold.colabfold import run_alphafold

run_alphafold is specifically in the colabfold module

Quickstart

Minimal example to run AlphaFold2 prediction using ColabFold patched pipeline. You must have AlphaFold databases and model parameters downloaded.

from alphafold.colabfold import run_alphafold
import os
# Set environment variables for required paths
os.environ['ALPHAFOLD_DB_DIR'] = '/path/to/databases'  # Adjust to your DB path
os.environ['ALPHAFOLD_MODEL_DIR'] = '/path/to/models'  # Adjust to your model params
# Basic usage
run_alphafold(
    fasta_path='sequence.fasta',
    output_dir='./output',
    num_models=1,
    model_names='model_1',
    use_amber=True,
    use_gpu_relax=True
)