ParaFold

Author: Bozitao Zhong - [email protected]

📑 Please cite our paper if you used ParaFold (ParallelFold) in you research.

Overview

Recent change: ParaFold now supports AlphaFold 2.3.1

This project is a modified version of DeepMind's AlphaFold2 to achieve high-throughput protein structure prediction.

We have these following modifications to the original AlphaFold pipeline:

• Divide CPU part (MSA and template searching) and GPU part (prediction model)

How to install

We recommend to install AlphaFold locally, and not using docker.

# clone this repo
git clone https://github.com/Zuricho/ParallelFold.git

# Create a miniconda environment for ParaFold/AlphaFold
# Recommend you to use python 3.8, version < 3.7 have missing packages, python versions newer than 3.8 were not tested
conda create -n parafold python=3.8

pip install py3dmol
# openmm 7.7 is recommended (original alphafold using 7.5.1, but it is not supported now)
conda install -c conda-forge openmm=7.7 pdbfixer

# use pip3 to install most of packages
pip3 install -r requirements.txt

# install cuda and cudnn
# cudatoolkit 11.3.1 matches cudnn 8.2.1
conda install cudatoolkit=11.3 cudnn

# downgrade jaxlib to the correct version, matches with cuda and cudnn version
pip3 install --upgrade --no-cache-dir jax==0.3.25 jaxlib==0.3.25+cuda11.cudnn82 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

# install packages for multiple sequence alignment
conda install -c bioconda hmmer=3.3.2 hhsuite=3.3.0 kalign2=2.04

chmod +x run_alphafold.sh

Download the sequence database

You can use the downloading script from AlphaFold repo. The data directory should have the following directory structure. Some old versions of AlphaFold might have older database versions, you should update them (reference to AlphaFold repo)

$DOWNLOAD_DIR/                             # Total: ~ 2.62 TB (download: 556 GB)
    bfd/                                   # ~ 1.8 TB (download: 271.6 GB)
        # 6 files.
    mgnify/                                # ~ 120 GB (download: 67 GB)
        mgy_clusters_2022_05.fa
    params/                                # ~ 5.3 GB (download: 5.3 GB)
        # 5 CASP14 models,
        # 5 pTM models,
        # 5 AlphaFold-Multimer models,
        # LICENSE,
        # = 16 files.
    pdb70/                                 # ~ 56 GB (download: 19.5 GB)
        # 9 files.
    pdb_mmcif/                             # ~ 238 GB (download: 43 GB)
        mmcif_files/
            # About 199,000 .cif files.
        obsolete.dat
    pdb_seqres/                            # ~ 0.2 GB (download: 0.2 GB)
        pdb_seqres.txt
    small_bfd/                             # ~ 17 GB (download: 9.6 GB)
        bfd-first_non_consensus_sequences.fasta
    uniref30/                              # ~ 206 GB (download: 52.5 GB)
        # 7 files.
    uniprot/                               # ~ 105 GB (download: 53 GB)
        uniprot.fasta
    uniref90/                              # ~ 67 GB (download: 34 GB)
        uniref90.fasta

Some detail information of modified files

• run_alphafold.py: modified version of original run_alphafold.py, it has multiple additional functions like skipping featuring steps when exists feature.pkl in output folder
• run_alphafold.sh: bash script to run run_alphafold.py

How to run

Run features

Run on CPUs to get features:

./run_alphafold.sh \
-d data \
-o output \
-p monomer_ptm \
-i input/GA98.fasta \
-t 1800-01-01 \
-m model_1 \
-f

-f means only run the featurization step, result in a feature.pkl file, and skip the following steps.

8 CPUs is enough, according to my test, more CPUs won't help with speed

Featuring step will output the feature.pkl and MSA folder in your output folder: ./output/[FASTA_NAME]/

PS: Here we put input files in an input folder to organize files in a better way.

Run monomer prediction

After the feature step, you can run run_alphafold.sh using GPU:

./run_alphafold.sh \
-d data \
-o output \
-m model_1,model_2,model_3,model_4,model_5 \
-p monomer_ptm \
-i input/GA98.fasta \
-t 1800-01-01 

If you have successfully output feature.pkl, you can have a very fast featuring step

Run multimer prediction

./run_alphafold.sh \
-d data \
-o output \
-m model_1_multimer,model_2_multimer,model_3_multimer,model_4_multimer,model_5_multimer \
-p multimer \
-i input/GA98.fasta \
-t 1800-01-01 

What is this for

ParallelFold can help you accelerate AlphaFold when you want to predict multiple sequences. After dividing the CPU part and GPU part, users can finish feature step by multiple processors. Using ParaFold, you can run AlphaFold 2~3 times faster than DeepMind's procedure.

If you have any question, please raise issues