AlphaFold2.3 conda版本详细安装与使用
文章目录
AlphaFold2.3 conda版本详细安装与使用前言一、安装环境二、安装步骤1. 确认已安装的cudatoolkit版本和cudnn版本2. 创建虚拟环境alphafold-env并激活，指定python=3.83.安装AlphaFold2.3支持的新版openmm，CUDA等4. clone alphafold并进入文件夹5. 下载数据库及模型参数文件（参考官方说明）6. 安装涉及到的依赖包。7. 验证环境是否安装成功8. 运行alphafold9. 对alphafold预测结构使用Pymol根据pLDDT染色
安装-使用总结
前言
（1）结合AlphaFold官方提供了Docker安装指引，本文提供了conda版本的安装，配合run_alphafold.sh脚本使用更加方便。 （2）根据AlphaFold官方python脚本，更新了run_alphafold.sh中的models_to_relax参数，便于AF2.3版本使用。 （3）对alphafold预测结果使用Pymol根据pLDDT染色。
一、安装环境
本例安装环境：ubuntu 22.04, CUDA版本11.8，cudnn版本 8.6.0，RTX3060，4T机械硬盘。
二、安装步骤
1. 确认已安装的cudatoolkit版本和cudnn版本
通过nvidia-smi命令查看CUDA driver版本，如图为12.2。 通过 nvcc -V命令查看CUDA runtime版本，如图为11.8。 确保Alphafold2安装CUDA版本不高于CUDA runtime版本，即≤11.8。 通过 dpkg -l | grep cudnn 查看cudnn版本，本安装cudnn版本为8.6.0。 Note：确保CUDA和cudnn的版本匹配，TensorFlow版本需要与CUDA匹配。
2. 创建虚拟环境alphafold-env并激活，指定python=3.8
conda create -n alphafold-env python=3.8
conda activate alphafold-env
3.安装AlphaFold2.3支持的新版openmm，CUDA等
conda install -c conda-forge openmm==7.7.0 cudatoolkit==11.8.0
conda install -c bioconda hmmer==3.3.2 hhsuite==3.3.0 kalign2==2.04
conda install pdbfixer mock
4. clone alphafold并进入文件夹
git clone https://kkgithub.com/deepmind/alphafold.git
cd ./alphafold
#下载残基性质参数到common文件夹
wget -q -P alphafold/common/ https://git.scicore.unibas.ch/schwede/openstructure/-/raw/7102c63615b64735c4941278d92b554ec94415f8/modules/mol/alg/src/stereo_chemical_props.txt
5. 下载数据库及模型参数文件（参考官方说明）
#安装下载软件
sudo apt install aria2
使用官方脚本，位于alphafold/scripts/
sh scripts/download_all_data.sh
NOTE: 保存数据库文件夹需要在alphafold主文件夹（即下文的 $data_dir）以外。 alphafold主文件夹可以位于任意位置，保持读写权限即可。 下载结束后，文件目录tree如下： 大小超过2.8T。
6. 安装涉及到的依赖包。
将如下内容保存到requirement.txt文件，使用pip安装。
absl-py==1.0.0
biopython==1.79
chex==0.0.7
dm-haiku==0.0.9
dm-tree==0.1.6
docker==5.0.0
immutabledict==2.0.0
ml-collections==0.1.0
pillow==10.1.0
numpy==1.21.6
pandas==1.3.4
scipy==1.7.3
tensorflow-gpu==2.11.0
zipp==3.17.0
jax==0.4.13
#根据安装情况调整
pip install -r requirement.txt
#安装jaxlib：与CUDA和cudnn硬件的版本可以查看https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
pip install --upgrade jaxlib==0.4.13+cuda11.cudnn86 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
检查tensorflow-gpu安装：python
import tensorflow as tf
tf.config.list_physical_devices('GPU')
显示有类似如下信息，即为正常。
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
7. 验证环境是否安装成功
cd alphafold
#在主文件夹运行：
python run_alphafold_test.py
如有提示缺少安装文件，用conda或者pip安装。最终提示OK即可，示意如下图。
8. 运行alphafold
（1）使用run_alphafold.py运行： 对于单链蛋白（monomer）搜索full_dbs。
conda activate alphafold-env
cd alphafold
python run_alphafold.py \
--fasta_paths=/path/to/input/query.fasta \
--output_dir=/path/to/output/full \
--uniref90_database_path=$data_dir/uniref90/uniref90.fasta \
--mgnify_database_path=$data_dir/mgnify/mgy_clusters_2022_05.fa \
--data_dir=$data_dir \
--template_mmcif_dir=$data_dir/pdb_mmcif/mmcif_files \
--obsolete_pdbs_path=$data_dir/pdb_mmcif/obsolete.dat \
--pdb70_database_path=$data_dir/pdb70/pdb70 \
--uniref30_database_path=$data_dir/uniref30/UniRef30_2021_03 \
--bfd_database_path=$data_dir/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt \
--max_template_date=2023-10-31 \
--models_to_relax=best \
--use_gpu_relax=true \
--db_preset=full_dbs \
--model_preset=monomer
（2）也可以使用run_alphafold.sh脚本便于输入，脚本从如下获取： https://kkgithub.com/kalininalab/alphafold_non_docker/blob/main/run_alphafold.sh 对于AF2.1以后的版本，run_relax 已被models_to_relax替代，为了顺利运行run_alphafold.sh，笔者对 -r 参数做了修改，保持默认对最优输出结构（5个输出中pLDDT最高的1个）做relax（models_to_relax=“best”）。 新脚本命名为run_alphafold23.sh，脚本内容、使用方法和脚本使用示例如下。
#!/bin/bash
# Description: AlphaFold non-docker version
# Author: Sanjay Kumar Srikakulam
# modifier: Blockbuater_drug according to AF v2.3
usage() {
echo ""
echo "Please make sure all required parameters are given"
echo "Usage: $0
echo "Required Parameters:"
echo "-d
echo "-o
echo "-f
echo "-t
echo "Optional Parameters:"
echo "-g
echo "-r
echo "-e
echo "-n
echo "-a
echo "-m
echo "-c
echo "-p
echo "-l
echo "-b
echo ""
exit 1
}
while getopts ":d:o:f:t:g:r:e:n:a:m:c:p:l:b:" i; do
case "${i}" in
d)
data_dir=$OPTARG
;;
o)
output_dir=$OPTARG
;;
f)
fasta_path=$OPTARG
;;
t)
max_template_date=$OPTARG
;;
g)
use_gpu=$OPTARG
;;
r)
models_to_relax=$OPTARG
;;
e)
enable_gpu_relax=$OPTARG
;;
n)
openmm_threads=$OPTARG
;;
a)
gpu_devices=$OPTARG
;;
m)
model_preset=$OPTARG
;;
c)
db_preset=$OPTARG
;;
p)
use_precomputed_msas=$OPTARG
;;
l)
num_multimer_predictions_per_model=$OPTARG
;;
b)
benchmark=$OPTARG
;;
esac
done
# Parse input and set defaults
if [[ "$data_dir" == "" || "$output_dir" == "" || "$fasta_path" == "" || "$max_template_date" == "" ]] ; then
usage
fi
if [[ "$benchmark" == "" ]] ; then
benchmark=false
fi
if [[ "$use_gpu" == "" ]] ; then
use_gpu=true
fi
if [[ "$gpu_devices" == "" ]] ; then
gpu_devices=0
fi
if [[ "$models_to_relax" == "" ]] ; then
models_to_relax="best"
fi
if [[ "$models_to_relax" != "best" && "$models_to_relax" != "all" && "$models_to_relax" != "none" ]] ; then
echo "Unknown models to relax preset! Using default ('best')"
db_preset="best"
fi
if [[ "$enable_gpu_relax" == "" ]] ; then
enable_gpu_relax="true"
fi
if [[ "$enable_gpu_relax" == true && "$use_gpu" == true ]] ; then
use_gpu_relax="true"
else
use_gpu_relax="false"
fi
if [[ "$num_multimer_predictions_per_model" == "" ]] ; then
num_multimer_predictions_per_model=5
fi
if [[ "$model_preset" == "" ]] ; then
model_preset="monomer"
fi
if [[ "$model_preset" != "monomer" && "$model_preset" != "monomer_casp14" && "$model_preset" != "monomer_ptm" && "$model_preset" != "multimer" ]] ; then
echo "Unknown model preset! Using default ('monomer')"
model_preset="monomer"
fi
if [[ "$db_preset" == "" ]] ; then
db_preset="full_dbs"
fi
if [[ "$db_preset" != "full_dbs" && "$db_preset" != "reduced_dbs" ]] ; then
echo "Unknown database preset! Using default ('full_dbs')"
db_preset="full_dbs"
fi
if [[ "$use_precomputed_msas" == "" ]] ; then
use_precomputed_msas="false"
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)
uniref90_database_path="$data_dir/uniref90/uniref90.fasta"
mgnify_database_path="$data_dir/mgnify/mgy_clusters_2022_05.fa"
template_mmcif_dir="$data_dir/pdb_mmcif/mmcif_files"
obsolete_pdbs_path="$data_dir/pdb_mmcif/obsolete.dat"
uniprot_database_path="$data_dir/uniprot/uniprot.fasta"
pdb_seqres_database_path="$data_dir/pdb_seqres/pdb_seqres.txt"
pdb70_database_path="$data_dir/pdb70/pdb70"
bfd_database_path="$data_dir/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt"
small_bfd_database_path="$data_dir/small_bfd/bfd-first_non_consensus_sequences.fasta"
uniref30_database_path="$data_dir/uniref30/UniRef30_2021_03"
# 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)
binary_paths="--hhblits_binary_path=$hhblits_binary_path --hhsearch_binary_path=$hhsearch_binary_path --jackhmmer_binary_path=$jackhmmer_binary_path --kalign_binary_path=$kalign_binary_path"
database_paths="--uniref90_database_path=$uniref90_database_path --mgnify_database_path=$mgnify_database_path --data_dir=$data_dir --template_mmcif_dir=$template_mmcif_dir --obsolete_pdbs_path=$obsolete_pdbs_path"
if [[ $model_preset == "multimer" ]]; then
database_paths="$database_paths --uniprot_database_path=$uniprot_database_path --pdb_seqres_database_path=$pdb_seqres_database_path"
else
database_paths="$database_paths --pdb70_database_path=$pdb70_database_path"
fi
if [[ "$db_preset" == "reduced_dbs" ]]; then
database_paths="$database_paths --small_bfd_database_path=$small_bfd_database_path"
else
database_paths="$database_paths --uniref30_database_path=$uniref30_database_path --bfd_database_path=$bfd_database_path"
fi
command_args="--fasta_paths=$fasta_path --output_dir=$output_dir --max_template_date=$max_template_date --db_preset=$db_preset --model_preset=$model_preset --benchmark=$benchmark --use_precomputed_msas=$use_precomputed_msas --num_multimer_predictions_per_model=$num_multimer_predictions_per_model --models_to_relax=$models_to_relax --use_gpu_relax=$use_gpu_relax --logtostderr"
# Run AlphaFold with required parameters
$(python $alphafold_script $binary_paths $database_paths $command_args)
run_alphafold23.sh使用方法：
Usage: run_alphafold23.sh
Required Parameters:
-d
-o
-f
-t
Optional Parameters:
-g
-r
-e
-n
-a
-m
-c
-p
-l
-b
run_alphafold23.sh使用示例： 将run_alphafold23.sh置于alphafold主目录。 下载rcsb_pdb_8I55.fasta文件（https://www.rcsb.org/fasta/entry/8I55）作为序列输入。 /path/to/inputdir和/path/to/outputdir分别为输入和输出文件夹路径。
conda activate alphafold-env
cd alphafold
bash ./run_alphafold23.sh -d $data_dir -o /path/to/outputdir -f /path/to/inputdir/rcsb_pdb_8I55.fasta -t 2023-10-31 -c reduced_dbs
运行~20min，结果文件夹中有以下输出文件： ranking_debug.json ：5种模型预测结构的pLDDT打分。
{
"plddts": {
"model_1_pred_0": 87.53338158106746,
"model_2_pred_0": 86.49349522466628,
"model_3_pred_0": 85.68943193089224,
"model_4_pred_0": 84.15560768736528,
"model_5_pred_0": 84.52065593375735
},
"order": [
"model_1_pred_0",
"model_2_pred_0",
"model_3_pred_0",
"model_5_pred_0",
"model_4_pred_0"
]
}
timings.json： 记录的运行时间如下，单位s。
{
"features": 1483.1720588207245,
"process_features_model_1_pred_0": 2.3148465156555176,
"predict_and_compile_model_1_pred_0": 85.52882599830627,
"process_features_model_2_pred_0": 1.143226146697998,
"predict_and_compile_model_2_pred_0": 80.66072988510132,
"process_features_model_3_pred_0": 1.0205607414245605,
"predict_and_compile_model_3_pred_0": 63.78041362762451,
"process_features_model_4_pred_0": 1.0224313735961914,
"predict_and_compile_model_4_pred_0": 63.5577335357666,
"process_features_model_5_pred_0": 1.0538318157196045,
"predict_and_compile_model_5_pred_0": 63.124531269073486,
"relax_model_1_pred_0": 5.963490009307861
}
confidence_model_1_pred_0.json ：每个残基的打分。
{"residueNumber":[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143],"confidenceScore":[93.52,97.19,98.5,98.72,98.51,98.37,97.37,95.5,91.07,88.46,88.55,89.87,91.54,91.29,90.86,89.49,94.56,96.18,94.89,96.9,97.53,97.97,97.87,98.19,98.32,98.35,97.94,97.83,98.04,97.64,96.6,97.23,97.59,97.23,98.33,98.62,98.37,98.3,96.92,95.2,91.22,87.58,88.01,87.34,91.07,95.15,94.87,95.57,95.05,96.04,97.44,97.35,96.61,97.26,98.01,97.39,96.96,97.79,97.62,98.04,97.87,98.31,98.5,98.54,98.3,97.84,97.28,96.61,94.93,94.55,93.86,91.7,89.53,85.18,85.64,86.04,87.81,89.16,90.58,92.12,92.65,91.3,90.9,92.83,90.74,89.27,87.36,75.99,64.56,57.37,51.47,39.82,47.69,40.84,47.46,35.5,47.18,44.85,41.38,48.78,42.01,51.79,49.66,51.94,51.03,48.19,51.79,52.1,57.6,64.85,71.12,84.97,88.06,92.0,94.13,93.41,95.49,97.16,97.29,97.29,97.52,97.3,97.12,95.08,95.7,95.48,93.76,94.57,95.46,97.16,96.88,96.5,96.89,97.32,96.75,97.27,98.05,97.98,96.94,97.82,97.79,94.74,87.27],"confidenceCategory":["H","H","H","H","H","H","H","H","H","M","M","M","H","H","H","M","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","M","M","M","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","M","M","M","M","M","M","H","H","H","H","H","H","H","M","M","M","L","L","L","D","D","D","D","D","D","D","D","D","D","L","D","L","L","D","L","L","L","L","M","M","M","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","H","M"]}
relaxed_model_1_pred_0.pdb：为relaxation之后的结构，相比增加了主链和侧链的氢原子，结构文件中第11列即为pLDDT，对应实验获得pdb文件中的b-factor一列。 relaxation阶段采用openmm amber_relaxer进行minimization，ff99sb力场，用到GPU。
9. 对alphafold预测结构使用Pymol根据pLDDT染色
Pymol安装可参考（ubuntu 安装 最新 PyMOL [源码安装][免费]）：https://blog.csdn.net/zdx1996/article/details/121656769 打开Pymol，file→Edit pymolrc，确定，打开pymolrc，输入如下内容；点击file→Save，关闭.pymolrc窗口。
run https://raw.kkgithub.com/cbalbin-bio/pymol-color-alphafold/master/coloraf.py
在Pymol 中使用命令：coloraf relaxed_model_1_pred_0 结果如下图，蓝色blue和青色cyan标识的区域分别为预测残基置信度pLDDT>90和90>pLDDT>70，属于可信度高的区域。 通过align，alphafold预测最优结果relaxation后，与X-ray pdb（8I55）结构中4个α-helix与4个β-sheet目视均预测良好，计算RMSD=0.35； 预测结构给出了loop可能的构象，相应这部分在X-ray结构中是缺失的，如下图。
安装-使用总结
（1）MSA阶段文件搜索较为耗时，硬盘读取速度较为重要；模型预测阶段minimization用到GPU，仅用CPU也可以，但耗时将显著增加；150AA蛋白Monomer搜索reduced_dbs，整个预测+relax在笔者硬件条件下大致需要25 min，同样条件full_dbs大致需要2 h。 （2）Multimer运行时间更久，30AA+150AA的双链Multimer，笔者硬件配置下预测时间超过了4.5 h；如要用来研究预测PPI，一定的硬件配置是必要的，尤其是硬盘和GPU。 以上就是AlphaFold2.3的conda环境安装及使用体验，欢迎批评指正与交流。
推荐阅读
您阅读本篇文章共花了：
发表评论