AlphaFold2配置和部署过程

AlphaFold2配置和部署过程

参考：https://github.com/kalininalab/alphafold_non_docker

1.Environment

CPU：Intel® Xeon® Gold 6240 CPU @ 2.60GHz
GPU：Tesla V100S PCIe 32GB * 4
OS：Ubuntu 20.04.2 LTS
CUDA：11.3
NVIDIA：NVIDIA driver version: 470.57.02

2.Insatll environment dependencies

cudnn和cudatoolkit根据自己电脑配置

conda install -y -c conda-forge openmm==7.5.1 cudnn==8.2.1.32 cudatoolkit==11.0.3 pdbfixer==1.7
conda install -y -c bioconda hmmer==3.3.2 hhsuite==3.3.0 kalign2==2.04

3.Download chemical properties to the common folder

wget -q -P alphafold/alphafold/common/ https://git.scicore.unibas.ch/schwede/openstructure/-/raw/7102c63615b64735c4941278d92b554ec94415f8/modules/mol/alg/src/stereo_chemical_props.txt

4.Install alphafold dependencies

pip install absl-py==0.13.0 biopython==1.79 chex==0.0.7 dm-haiku==0.0.4 dm-tree==0.1.6 immutabledict==2.0.0 jax==0.2.14 ml-collections==0.1.0 numpy==1.19.5 scipy==1.7.0 tensorflow==2.5.0
pip install --upgrade jax jaxlib==0.1.69+cuda111 -f https://storage.googleapis.com/jax-releases/jax_releases.html

注意点：
1.jaxlib版本要下载正确，与cuda的版本一致，不然在run alphafold的时候不能调用GPU。
2.jaxlib只能在linux下使用，windows下没有这个包，没有linux建议放弃。

5.Apply OpenMM patch

cd ~/anaconda3/envs/alphafold/lib/python3.8/site-packages/ 
patch -p0 < $alphafold_path/docker/openmm.patch

其中$alphafold_path表示自己电脑上alphafold存放的位置

6.Download all databases

跟着官方的下载渠道走： https://github.com/deepmind/alphafold#genetic-databases
解压后的数据大约2.2T，存放在SSD上效果更佳。

7.Run AlphaFold2

两种选择：
1.在源码的run_alphafold.py中将参数中各个数据库的路径、输出结果的路径，以及其他参数（例如GPU设置、需要预测的蛋白质氨基酸序列 [存放在fastashang] ）进行配置。
2.写一个script,用于将各个数据库的路径拼接,配置相关的一些参数，以及run run_alphafold.py,具体的如下(该来自参考的作者)

#!/bin/bash
# Description: AlphaFold non-docker version
# Author: Sanjay Kumar Srikakulam

usage() {
        echo ""
        echo "Please make sure all required parameters are given"
        echo "Usage: $0 <OPTIONS>"
        echo "Required Parameters:"
        echo "-d <data_dir>     Path to directory of supporting data"
        echo "-o <output_dir>   Path to a directory that will store the results."
        echo "-m <model_names>  Names of models to use (a comma separated list)"
        echo "-f <fasta_path>   Path to a FASTA file containing one sequence"
        echo "-t <max_template_date> Maximum template release date to consider (ISO-8601 format - i.e. YYYY-MM-DD). Important if folding historical test sets"
        echo "Optional Parameters:"
        echo "-n <openmm_threads>   OpenMM threads (default: all available cores)"
        echo "-b <benchmark>    Run multiple JAX model evaluations to obtain a timing that excludes the compilation time, which should be more indicative of the time required for inferencing many proteins (default: 'False')"
        echo "-g <use_gpu>      Enable NVIDIA runtime to run with GPUs (default: True)"
        echo "-a <gpu_devices>  Comma separated list of devices to pass to 'CUDA_VISIBLE_DEVICES' (default: 0)"
        echo "-p <preset>       Choose preset model configuration - no ensembling and smaller genetic database config (reduced_dbs), no ensembling and full genetic database config  (full_dbs) or full genetic database config and 8 model ensemblings (casp14)"
        echo ""
        exit 1
}

while getopts ":d:o:m:f:t:g:n:a:p:b" i; do
        case "${i}" in
        d)
                data_dir=$OPTARG
        ;;
        o)
                output_dir=$OPTARG
        ;;
        m)
                model_names=$OPTARG
        ;;
        f)
                fasta_path=$OPTARG
        ;;
        t)
                max_template_date=$OPTARG
        ;;
        g)
                use_gpu=$OPTARG
        ;;
        n)
                openmm_threads=$OPTARG
        ;;
        a)
                gpu_devices=$OPTARG
        ;;
        p)
                preset=$OPTARG
        ;;
        b)
                benchmark=true
        ;;
        esac
done

# Parse input and set defaults
if [[ "$data_dir" == "" || "$output_dir" == "" || "$model_names" == "" || "$fasta_path" == "" || "$max_template_date" == "" ]] ; then
    usage
fi

if [[ "$benchmark" == "" ]] ; then
    benchmark=false
fi

if [[ "$use_gpu" == "" ]] ; then
    use_gpu=true
fi

if [[ "$gpu_devices" == "" ]] ; then
    gpu_devices=0
fi

if [[ "$preset" == "" ]] ; then
    preset="full_dbs"
fi

if [[ "$preset" != "full_dbs" && "$preset" != "casp14" && "$preset" != "reduced_dbs" ]] ; then
    echo "Unknown preset! Using default ('full_dbs')"
    preset="full_dbs"
fi

# This bash script looks for the run_alphafold.py script in its current working directory, if it does not exist then exits
current_working_dir=$(pwd)
alphafold_script="$current_working_dir/run_alphafold.py"

if [ ! -f "$alphafold_script" ]; then
    echo "Alphafold python script $alphafold_script does not exist."
    exit 1
fi

# Export ENVIRONMENT variables and set CUDA devices for use
# CUDA GPU control
export CUDA_VISIBLE_DEVICES=-1
if [[ "$use_gpu" == true ]] ; then
    export CUDA_VISIBLE_DEVICES=0

    if [[ "$gpu_devices" ]] ; then
        export CUDA_VISIBLE_DEVICES=$gpu_devices
    fi
fi

# OpenMM threads control
if [[ "$openmm_threads" ]] ; then
    export OPENMM_CPU_THREADS=$openmm_threads
fi

# TensorFlow control
export TF_FORCE_UNIFIED_MEMORY='1'

# JAX control
export XLA_PYTHON_CLIENT_MEM_FRACTION='4.0'

# Path and user config (change me if required)
bfd_database_path="$data_dir/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt"
small_bfd_database_path="$data_dir/small_bfd/bfd-first_non_consensus_sequences.fasta"
mgnify_database_path="$data_dir/mgnify/mgy_clusters.fa"
template_mmcif_dir="$data_dir/pdb_mmcif/mmcif_files"
obsolete_pdbs_path="$data_dir/pdb_mmcif/obsolete.dat"
pdb70_database_path="$data_dir/pdb70/pdb70"
uniclust30_database_path="$data_dir/uniclust30/uniclust30_2018_08/uniclust30_2018_08"
uniref90_database_path="$data_dir/uniref90/uniref90.fasta"

# Binary path (change me if required)
hhblits_binary_path=$(which hhblits)
hhsearch_binary_path=$(which hhsearch)
jackhmmer_binary_path=$(which jackhmmer)
kalign_binary_path=$(which kalign)

# Run AlphaFold with required parameters
# 'reduced_dbs' preset does not use bfd and uniclust30 databases
if [[ "$preset" == "reduced_dbs" ]]; then
    $(python $alphafold_script --hhblits_binary_path=$hhblits_binary_path --hhsearch_binary_path=$hhsearch_binary_path --jackhmmer_binary_path=$jackhmmer_binary_path --kalign_binary_path=$kalign_binary_path --small_bfd_database_path=$small_bfd_database_path --mgnify_database_path=$mgnify_database_path --template_mmcif_dir=$template_mmcif_dir --obsolete_pdbs_path=$obsolete_pdbs_path --pdb70_database_path=$pdb70_database_path --uniref90_database_path=$uniref90_database_path --data_dir=$data_dir --output_dir=$output_dir --fasta_paths=$fasta_path --model_names=$model_names --max_template_date=$max_template_date --preset=$preset --benchmark=$benchmark --logtostderr)
else
    $(python $alphafold_script --hhblits_binary_path=$hhblits_binary_path --hhsearch_binary_path=$hhsearch_binary_path --jackhmmer_binary_path=$jackhmmer_binary_path --kalign_binary_path=$kalign_binary_path --bfd_database_path=$bfd_database_path --mgnify_database_path=$mgnify_database_path --template_mmcif_dir=$template_mmcif_dir --obsolete_pdbs_path=$obsolete_pdbs_path --pdb70_database_path=$pdb70_database_path --uniclust30_database_path=$uniclust30_database_path --uniref90_database_path=$uniref90_database_path --data_dir=$data_dir --output_dir=$output_dir --fasta_paths=$fasta_path --model_names=$model_names --max_template_date=$max_template_date --preset=$preset --benchmark=$benchmark --logtostderr)
fi

如果在源码上修改，则直接 python run_alphafold.py
如果使用bash script,则可以使用以下命令,-d 后是数据库存放的位置，-o 是结果输出的路径，-m 是选择模型（可以选多个）， -f 是查询序列方法的路径，-t 是使用该时间之前的数据库内容。

# Example run (Uses the GPU with index id 0 as default)
bash run_alphafold.sh -d ./alphafold_data/ -o ./outputs/ -m model_1 -f ./xxx/T100.fasta -t 2020-05-14

# OR for CPU only run
bash run_alphafold.sh -d ./alphafold_data/ -o ./outputs/ -m model_1 -f ./xxx/T100.fasta -t 2020-05-14 -g False

8.View protein 3D structure

1.安装生物大分子展示软件PyMOL，具体参考：http://pymol.chenzhaoqiang.com/intro/overview.html
我使用的是alphafold2作者在colab上 Single sequence input (no MSA)使用的序列，如下：

query_sequence = "GWSTELEKHREELKEFLKKEGITNVEIRIDNGRLEVRVEGGTERLKRFLEELRQKLEKKGYTVDIKIE"

在运行代码后在outputs里包含多个文件，现在后缀名为pdb的文件在PyMOL中打开，可以看到如下：

可以在pdb数据库中查找一些已有蛋白质三维结构的氨基酸序列进行验证。