参考资料:
1. https://zhuanlan.zhihu.com/p/7033616610AlphaFold3开源版的安装过程整理
2. rocky8安装alphafold3
部署过程
1. 升级显卡驱动为12.6 (NVIDIA GeForce RTX 4090)
2. 键入以下命令行
conda create -n AF3 python=3.12
conda activate AF3
git clone https://github.com/google-deepmind/alphafold3.git
cd alphafold3
mkdir ./hmmer_build ./hmmer
wget http://eddylab.org/software/hmmer/hmmer-3.4.tar.gz --directory-prefix ./hmmer_build
cd ./hmmer_build && tar zxf hmmer-3.4.tar.gz && rm hmmer-3.4.tar.gz
cd ./hmmer-3.4 | ./configure --prefix $(realpath ../../hmmer)
make -j8
make install
cd ./easel && make install
cd ../../../
rm -rf ./hmmer_build
pip3 install -r dev-requirements.txt
pip3 install --no-deps .
# failed in build pybind11
pip install pybind11
# try again
build_data
python run_alphafold.py --helpfull #测试
使用方法
AlphaFold 3 | e-Science Document 南京大学
AlphaFold 3 Input
AlphaFold 3 Output
cat > run_af3.sh << EOF
#! /bin/bash -e
# 用法说明
if [ $# -ne 2 ]; then
echo "usage: $0 <json_path> <output_dir>"
exit 1
fi
af3_run_dir="/home/sxwen/software/program/alphafold3"
database_dir="/home/sxwen/software/AlphaFold3/AF3_databases"
model_parameters_dir="/home/sxwen/software/AlphaFold3/AF3_parameters"
json_path="$1"
output_dir="$2"
# 检查必要目录是否存在
for dir in "$af3_run_dir" "$database_dir" "$model_parameters_dir"; do
if [ ! -d "$dir" ]; then
echo "Error: Directory don't exist: $dir"
exit 1
fi
done
# 创建输出目录
mkdir -p "$output_dir"
# 运行Alphafold3
python ${af3_run_dir}/run_alphafold.py \
--model_dir=${model_parameters_dir} \
--db_dir=${database_dir} \
--json_path=${json_path} \
--output_dir=${output_dir}
EOF
加速计算:使用colabfold生成AlphaFold3-compatible JSON format
With colabfold_search:
colabfold_search --mmseqs /path/to/bin/mmseqs input_sequences.fasta /path/to/db_folder msas --af3-json
With colabfold_batch:
colabfold_batch input_sequences.fasta out_dir --af3-json
运行AF3:
chmod +x run_af3.sh
./run_af3.sh input_sequences.json /path/to/save/result
做个测试吧
以转氨酶3FCR为例,生成dimer
绿色(晶体结构) VS 肉色(AF3预测)。结果还是不错的。