Open zhoujingyu13687306871 opened 1 year ago
hello,I am very happy to see that fastfold has added the multimer function, but I have a problem. When using the monomer function, I still cannot predict 5 models at once. Is there a solution for this? here is my scripts:
######################查看gpu利用率################################################ STATEFILE="state${BATCH_JOB_ID}" /usr/bin/touch ${STATE_FILE} function gpus_collection(){ while [[ cat "${STATE_FILE}" | grep "over" | wc -l == "0" ]]; do /usr/bin/sleep 1 /usr/bin/nvidia-smi >> "gpu_${BATCH_JOB_ID}.log" done } gpus_collection & #####################AF2计算部分################################################### module load anaconda/2021.11 module load cuda/11.3.0-gcc-4.8.5-oaa module load gcc/9.3.0-gcc-4.8.5-bxl source activate fastfold af2Root=/home/bingxing2/public
cat "${STATE_FILE}" | grep "over" | wc -l
python inference.py mono.fasta $af2Root/alphafold2.2.0/pdb_mmcif/mmcif_files \ --output_dir ./mono_out \ --uniref90_database_path $af2Root/uniref90/uniref90.fasta \ --mgnify_database_path $af2Root/mgnify/mgy_clusters.fa \ --pdb70_database_path $af2Root/pdb70/pdb70 \ --param_path $af2Root/alphafold2.2.0/params/params_model_1.npz \ --uniclust30_database_path $af2Root/uniclust30/uniclust30_2018_08/uniclust30_2018_08 \ --bfd_database_path $af2Root/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt \ --jackhmmer_binary_path which jackhmmer \ --hhblits_binary_path which hhblits \ --hhsearch_binary_path which hhsearch \ --kalign_binary_path which kalign \ --gpus 2 \ --enable_workflow \ --chunk_size 1 \ --inplace echo "over" >> "${STATE_FILE}"
which jackhmmer
which hhblits
which hhsearch
which kalign
Our inference scripts can only inference one model at a time. I think you can specify --model_name to use the other 4 model.
--model_name
hello,I am very happy to see that fastfold has added the multimer function, but I have a problem. When using the monomer function, I still cannot predict 5 models at once. Is there a solution for this? here is my scripts:
!/bin/bash
DSUB --job_type cosched
DSUB -n fastfold
DSUB -A root.bingxing2.gpuuser001
DSUB -q root.default
DSUB -R 'cpu=12;gpu=2;mem=90000'
DSUB -l wuhanG5500
DSUB -N 1
DSUB -e %J.out
DSUB -o %J.out
######################查看gpu利用率################################################ STATEFILE="state${BATCH_JOB_ID}" /usr/bin/touch ${STATE_FILE} function gpus_collection(){ while [[
cat "${STATE_FILE}" | grep "over" | wc -l
== "0" ]]; do /usr/bin/sleep 1 /usr/bin/nvidia-smi >> "gpu_${BATCH_JOB_ID}.log" done } gpus_collection & #####################AF2计算部分################################################### module load anaconda/2021.11 module load cuda/11.3.0-gcc-4.8.5-oaa module load gcc/9.3.0-gcc-4.8.5-bxl source activate fastfold af2Root=/home/bingxing2/publicadd '--gpus [N]' to use N gpus for inference
add '--enable_workflow' to use parallel workflow for data processing
add '--use_precomputed_alignments [path_to_alignments]' to use precomputed msa
add '--chunk_size [N]' to use chunk to reduce peak memory
add '--inplace' to use inplace to save memory
python inference.py mono.fasta $af2Root/alphafold2.2.0/pdb_mmcif/mmcif_files \ --output_dir ./mono_out \ --uniref90_database_path $af2Root/uniref90/uniref90.fasta \ --mgnify_database_path $af2Root/mgnify/mgy_clusters.fa \ --pdb70_database_path $af2Root/pdb70/pdb70 \ --param_path $af2Root/alphafold2.2.0/params/params_model_1.npz \ --uniclust30_database_path $af2Root/uniclust30/uniclust30_2018_08/uniclust30_2018_08 \ --bfd_database_path $af2Root/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt \ --jackhmmer_binary_path
which jackhmmer
\ --hhblits_binary_pathwhich hhblits
\ --hhsearch_binary_pathwhich hhsearch
\ --kalign_binary_pathwhich kalign
\ --gpus 2 \ --enable_workflow \ --chunk_size 1 \ --inplace echo "over" >> "${STATE_FILE}"