AlphaFold2 Benchmark ~ タンパク質の立体構造予測プログラム

タンパク質の立体構造予測プログラム AlphaFold2 (https://github.com/google-deepmind/alphafold)の ベンチマーク 結果を掲載します。AlphaFold2 は、DeepMind社が開発したタンパク質の立体構造予測プログラムであり、アミノ酸配列からタンパク質の立体構造を高い精度で予測することができます。また、GPU を利用することで高速に処理することが可能です。

テストした入力ファイル

  1. T1050 A7LXT1, Bacteroides Ovatus, 779 residues
  2. NP_001362351.1 multiple PDZ domain protein isoform 6 [Homo sapiens] 2000 aa
  3. KAG8491920.1 hypothetical protein CXB51_015235 [Gossypium anomalum] 2000 aa

T1050 A7LXT1, Bacteroides Ovatus, 779 residues

>T1050 A7LXT1, Bacteroides Ovatus, 779 residues|
MASQSYLFKHLEVSDGLSNNSVNTIYKDRDGFMWFGTTTGLNRYDGYTFKIYQHAENEPGSLPDNYITDIVEMPDGRFWINTARGYVLFDKERDYFITDVTGFMKNLESWGVPEQVFVDREGNTWLSVAGEGCYRYKEGGKRLFFSYTEHSLPEYGVTQMAECSDGILLIYNTGLLVCLDRATLAIKWQSDEIKKYIPGGKTIELSLFVDRDNCIWAYSLMGIWAYDCGTKSWRTDLTGIWSSRPDVIIHAVAQDIEGRIWVGKDYDGIDVLEKETGKVTSLVAHDDNGRSLPHNTIYDLYADRDGVMWVGTYKKGVSYYSESIFKFNMYEWGDITCIEQADEDRLWLGTNDHGILLWNRSTGKAEPFWRDAEGQLPNPVVSMLKSKDGKLWVGTFNGGLYCMNGSQVRSYKEGTGNALASNNVWALVEDDKGRIWIASLGGGLQCLEPLSGTFETYTSNNSALLENNVTSLCWVDDNTLFFGTASQGVGTMDMRTREIKKIQGQSDSMKLSNDAVNHVYKDSRGLVWIATREGLNVYDTRRHMFLDLFPVVEAKGNFIAAITEDQERNMWVSTSRKVIRVTVASDGKGSYLFDSRAYNSEDGLQNCDFNQRSIKTLHNGIIAIGGLYGVNIFAPDHIRYNKMLPNVMFTGLSLFDEAVKVGQSYGGRVLIEKELNDVENVEFDYKQNIFSVSFASDNYNLPEKTQYMYKLEGFNNDWLTLPVGVHNVTFTNLAPGKYVLRVKAINSDGYVGIKEATLGIVVNPPFKLAAALQHHHHHH

ジョブの実行方法

python3 docker/run_docker.py \
  –fasta_paths=T1050.fasta \
  –model_preset=monomer \
  –max_template_date=2024-04-01 \
  –data_dir=[データファイルのディレクトリ] \
  –output_dir=[出力結果先] \
–benchmark

ベンチマーク結果

ベンチマークNo.2
機器の特徴 (4) Intel Xeon Gold 6448H 2.4GHz 32C/64T + (4) NVIDIA RTX A6000
ベンチマーク実施日2024年4月27日
実行環境詳細HPC-ProServer DPeR960
CPU : (4) Intel Xeon Gold 6448H 2.4GHz 32C/64T
Mem : 4096GB (64) 64GB DDR5 4800MHz
SSD : 960GB SATA SSD
GPU : (4) NVIDIA RTX A6000
OS : RockyLinux 8.9
NVIDIA Driver : 550.54.15 , CUDA 12.4
Docker version 26.0.1, build d260a54
データは、25GbpsネットワークのNFS領域に配置
AlphaFold Version2.3.2
timeコマンド0.13user 0.01system 1:24:08elapsed 0%CPU (0avgtext+0avgdata 26880maxresident)k
808inputs+0outputs (3major+5042minor)pagefaults 0swaps
timings.json{
“features”: 3951.729302883148,
“process_features_model_1_pred_0”: 6.498595237731934,
“predict_and_compile_model_1_pred_0”: 232.60038781166077,
“process_features_model_2_pred_0”: 4.694603443145752,
“predict_and_compile_model_2_pred_0”: 204.78236889839172,
“process_features_model_3_pred_0”: 4.037086009979248,
“predict_and_compile_model_3_pred_0”: 203.1477324962616,
“process_features_model_4_pred_0”: 4.555547714233398,
“predict_and_compile_model_4_pred_0”: 199.04012203216553,
“process_features_model_5_pred_0”: 4.7832207679748535,
“predict_and_compile_model_5_pred_0”: 181.64868569374084,
“relax_model_2_pred_0”: 19.424469232559204
}
備考features : 3951.729302883148 sec , features以外 : 1065.21281933784 sec
GPUメモリ:約9GB使用
ベンチマークNo.1
機器の特徴 (2) Intel Xeon Gold 6430 2.1GHz 32C/64T + (2) NVIDIA RTX 6000 Ada
ベンチマーク実施日2024年4月27日
実行環境詳細HPC-ProServer DPrR7960
CPU : (2) Intel Xeon Gold 6430 2.1GHz 32C/64T
Mem : 512GB (16) 32GB DDR5 4800MHz
SSD : 1TB M.2 NVMe
GPU : (2) NVIDIA RTX 6000 Ada
OS : RockyLinux 8.9
NVIDIA Driver : 550.54.15 , CUDA 12.4
Docker version 26.0.1, build d260a54
データは、25GbpsネットワークのNFS領域に配置
AlphaFold Version2.3.2
timeコマンド0.15user 0.01system 1:30:43elapsed 0%CPU (0avgtext+0avgdata 27152maxresident)k
0inputs+0outputs (0major+5101minor)pagefaults 0swaps
timings.json{
“features”: 4743.83895778656,
“process_features_model_1_pred_0”: 5.070164918899536,
“predict_and_compile_model_1_pred_0”: 146.5927164554596,
“process_features_model_2_pred_0”: 4.10901141166687,
“predict_and_compile_model_2_pred_0”: 130.53641939163208,
“process_features_model_3_pred_0”: 3.5964343547821045,
“predict_and_compile_model_3_pred_0”: 120.6664125919342,
“process_features_model_4_pred_0”: 3.591764450073242,
“predict_and_compile_model_4_pred_0”: 116.03167223930359,
“process_features_model_5_pred_0”: 3.603001594543457,
“predict_and_compile_model_5_pred_0”: 111.4774022102356,
“relax_model_2_pred_0”: 22.427526473999023
}
備考features : 4743.838958 sec , features以外 : 667.7025261 sec
GPUメモリ:約9GB使用

NP_001362351.1 multiple PDZ domain protein isoform 6 [Homo sapiens] 2000 aa

>NP_001362351.1 multiple PDZ domain protein isoform 6 [Homo sapiens]
MLEAIDKNRALHAAERLQTKLRERGDVANEDKLSLLKSVLQSPLFSQILSLQTSVQQLKDQVNIATSATS
NIEYAHVPHLSPAVIPTLQNESFLLSPNNGNLEALTGPGIPHINGKPACDEFDQLIKNMAQGRHVEVFEL
LKPPSGGLGFSVVGLRSENRGELGIFVQEIQEGSVAHRDGRLKETDQILAINGQALDQTITHQQAISILQ
KAKDTVQLVIARGSLPQLVSPIVSRSPSAASTISAHSNPVHWQHMETIELVNDGSGLGFGIIGGKATGVI
VKTILPGGVADQHGRLCSGDHILKIGDTDLAGMSSEQVAQVLRQCGNRVKLMIARGAIEERTAPTALGIT
LSSSPTSTPELRVDASTQKGEESETFDVELTKNVQGLGITIAGYIGDKKLEPSGIFVKSITKSSAVEHDG
RIQIGDQIIAVDGTNLQGFTNQQAVEVLRHTGQTVLLTLMRRGMKQEAELMSREDVTKDADLSPVNASII
KENYEKDEDFLSSTRNTNILPTEEEGYPLLSAEIEEIEDAQKQEAALLTKWQRIMGINYEIVVAHVSKFS
ENSGLGISLEATVGHHFIRSVLPEGPVGHSGKLFSGDELLEVNGITLLGENHQDVVNILKELPIEVTMVC
CRRTVPPTTQSELDSLDLCDIELTEKPHVDLGEFIGSSETEDPVLAMTDAGQSTEEVQAPLAMWEAGIQH
IELEKGSKGLGFSILDYQDPIDPASTVIIIRSLVPGGIAEKDGRLLPGDRLMFVNDVNLENSSLEEAVEA
LKGAPSGTVRIGVAKPLPLSPEEGYVSAKEDSFLYPPHSCEEAGLADKPLFRADLALVGTNDADLVDEST
FESPYSPENDSIYSTQASILSLHGSSCGDGLNYGSSLPSSPPKDVIENSCDPVLDLHMSLEELYTQNLLQ
RQDENTPSVDISMGPASGFTINDYTPANAIEQQYECENTIVWTESHLPSEVISSAELPSVLPDSAGKGSE
YLLEQSSLACNAECVMLQNVSKESFERTINIAKGNSSLGMTVSANKDGLGMIVRSIIHGGAISRDGRIAI
GDCILSINEESTISVTNAQARAMLRRHSLIGPDIKITYVPAEHLEEFKISLGQQSGRVMALDIFSSYTGR
DIPELPEREEGEGEESELQNTAYSNWNQPRRVELWREPSKSLGISIVGGRGMGSRLSNGEVMRGIFIKHV
LEDSPAGKNGTLKPGDRIVEAPSQSESEPEKAPLCSVPPPPPSAFAEMGSDHTQSSASKISQDVDKEDEF
GYSWKNIRERYGTLTGELHMIELEKGHSGLGLSLAGNKDRSRMSVFIVGIDPNGAAGKDGRLQIADELLE
INGQILYGRSHQNASSIIKCAPSKVKIIFIRNKDAVNQMAVCPGNAVEPLPSNSENLQNKETEPTVTTSD
AAVDLSSFKNVQHLELPKDQGGLGIAISEEDTLSGVIIKSLTEHGVAATDGRLKVGDQILAVDDEIVVGY
PIEKFISLLKTAKMTVKLTIHAENPDSQAVPSAAGAASGEKKNSSQSLMVPQSGSPEPESIRNTSRSSTP
AIFASDPATCPIIPGCETTIEISKGRTGLGLSIVGGSDTLLGAIIIHEVYEEGAACKDGRLWAGDQILEV
NGIDLRKATHDEAINVLRQTPQRVRLTLYRDEAPYKEEEVCDTLTIELQKKPGKGLGLSIVGKRNDTGVF
VSDIVKGGIADADGRLMQGDQILMVNGEDVRNATQEAVAALLKCSLGTVTLEVGRIKAGPFHSERRPSQS
SQVSEGSLSSFTFPLSGSSTSESLESSSKKNALASEIQGLRTVEMKKGPTDSLGISIAGGVGSPLGDVPI
FIAMMHPTGVAAQTQKLRVGDRIVTICGTSTEGMTHTQAVNLLKNASGSIEMQVVAGGDVSVVTGHQQEP
ASSSLSFTGLTSSSIFQDDLGPPQCKSITLERGPDGLGFSIVGGYGSPHGDLPIYVKTVFAKGAASEDGR
LKRGDQIIAVNGQSLEGVTHEEAVAILKRTKGTVTLMVLS

ジョブの実行方法

python3 docker/run_docker.py \
  –fasta_paths=MPDZ.fasta \
  –model_preset=monomer \
  –max_template_date=2024-04-01 \
  –data_dir=[データファイルのディレクトリ] \
  –output_dir=[出力結果先] \
–benchmark=true

ベンチマーク結果

ベンチマークNo.3
機器の特徴 (2) Intel Xeon Gold 6426Y 2.5GHz 16C/32T + (2) NVIDIA RTX A6000
ベンチマーク実施日2024年6月16日
実行環境詳細HPC-ProServer DPrR7960
CPU : (4) Intel Xeon Gold 6426Y 2.5GHz 16C/32T
Mem : 256GB (8) 32GB DDR5 4800MHz
SSD(sys) : 1TB NVMe SSD
SSD(data) : 15.36TB U.2 NVMe SSD <= データはこの領域に配置
GPU : (2) NVIDIA RTX A6000
OS : RockyLinux 8.10
NVIDIA Driver : 555.42.02
Docker version 26.1.3, build b72abbb
AlphaFold Version2.3.2
timeコマンド0.14user 0.01system 4:46:14elapsed 0%CPU (0avgtext+0avgdata 26736maxresident)k
3784inputs+792outputs (28major+4631minor)pagefaults 0swaps
timings.json{
“features”: 4821.1200931072235,
“process_features_model_1_pred_0”: 17.36833930015564,
“predict_and_compile_model_1_pred_0”: 1369.2554149627686,
“predict_benchmark_model_1_pred_0”: 1302.739129781723,
“process_features_model_2_pred_0”: 16.627148389816284,
“predict_and_compile_model_2_pred_0”: 1242.8643596172333,
“predict_benchmark_model_2_pred_0”: 1194.1714398860931,
“process_features_model_3_pred_0”: 14.082069396972656,
“predict_and_compile_model_3_pred_0”: 1218.4357559680939,
“predict_benchmark_model_3_pred_0”: 1180.7467677593231,
“process_features_model_4_pred_0”: 14.319541692733765,
“predict_and_compile_model_4_pred_0”: 1214.740605354309,
“predict_benchmark_model_4_pred_0”: 1180.8296961784363,
“process_features_model_5_pred_0”: 13.722816944122314,
“predict_and_compile_model_5_pred_0”: 1109.9322047233582,
“predict_benchmark_model_5_pred_0”: 1074.2233765125275,
“relax_model_1_pred_0”: 151.10685849189758
}
備考features : 4821.1200931072235 sec , features以外 : 12315.16552 sec
GPUメモリ:約29GB使用

KAG8491920.1 hypothetical protein CXB51_015235 [Gossypium anomalum] 2000 aa

>KAG8491920.1 hypothetical protein CXB51_015235 [Gossypium anomalum]
MSNTCSYHHQKTCLYWYLTLVFCFFSPFSVKSNELQILLDLKSALNKSTTTAFNSWQTPNSICTFNGITC
NHEGFITELDLSTQNLTGILPFDSLCKLPSLQKLSFGYNSLHGPITGELNNCVKLQYLDLGNNFFTGFFP
NISSLIQLKFLHLNKSGFSGKFPWKSLENFTDLAVLSIGDNPFDRFQFPDQIFKLKKLNWLYMANCCIEG
KIPSAIGDLIELINLELENNYLSGDIPMEISKLHNLWQLELYYNNLTGKLPVGLRNLTKLEFFDASANKL
EGNISEMGYLNNLVSLHLYQNKFTGEIPPEFGQFRKLVNLSLYENMLTGPLPENLGSWANFDYIDVSENS
LTGPIPPYMCKQGTMRGLLLVQNRFTGEIPASYGNCKTLKRFRVNNNSLSGVVPAGIWGLPMVDIIDIAY
NRFEGPITSDIKNAKVMSILSVGFNRLSGELPQEISKAISLVKIEVNDNKFSGKIPHGIGELKRLNILKF
HNNMLSGSIPESLCSCVSLSDINMAVNSLSGKIPSCLGSLATLNSLNLSLNELSGKIPESLSSLKLNLFD
LSYNRLAGPIPESLSIEAYNGSLVGNPGLCSSTDRSFKRCQMGSGMSKDVHTIIVCFVIGVTVLLVSIGC
FVYLKRTEKDKNGDAHSLKEESWDVKSFHVLTFTEDEILDSIKQENLIGKGGSGNVYKVMLSNRVELAVK
HIWNTKSNSRRKTRSSAPMLTKHDGKAKELEAEVRTLSSIRHVNVVKLYCSITSEDSSLLVYEYLPNGSL
WDRLHTSKKMELDWDTRYEIAIGAAKGLEYLHHGCEKPVLHRDVKSSNILLDEYLKPKISDFGLAKIVQA
TSSMGNDSTHVIAGTHGYIAPEYGYTCKVDEKSDVYSFGVVLLELVTGKKPIEQEYGENKDIVSWVGSNL
KGKESVLSIVDPKIPHAFKEDAMKVLKIAILCTTTLPALRPTMRRVVQMLKEAEPYRLVGIVIVSFELRS
FKQESRRNRIGAESHATLGRGGLEHRDKRPHGDELGEARRGDVAFLAAKQGCDVMMSKKVTSRQGDLLKM
APKAVFLFLMLLSFMFYSSKAIRQNQWQFFSIMKASLSGNPLSDWEVNEGASYCNFTGVSCNNEGYVESM
NFSGWSLAGNFPADVCSYLPELRVLDISRNNFRGNFLNGVVNCSVLEVFNMSSVYLNATFPDLSKTTSLR
VLDLSYNRFRGDFSMSITNLTNLEVLYINENDGLNLWQLPTNISRLTKLRIMVLTTCKLYGRIPASIGNM
TSLVDLELSGNFLSGQIPKELGLLKNLQQLELYYNQHLSGTIPEELGNLIELIDLDMSVNHLSGSIPTSI
CRLPKLQVLQLYNNSLTGEIPGVIANSTTLTTLSLYGNFLSGQVPQNLGQLSPMVILDLSENQLSGSLPT
EVCRGGKLLYLLVLDNKLSGKLPDSYADCESLVRFRVSNNYLEGPIPEGLLGLPHVSIIDLADNNFTGHF
PGSIGNARNLSELFMQNNKVSGAIPRKISRAINMVKIDLSNNLLSGSIPTEIGNLKKLNLLVLQGNKLSS
SIPNSLSLLKSLNVLDLSSNCLTGNIPESLSELLPNFINLSNNNLSGPIPLSLIEGGLMESFSGNPGLCA
TVHIRNFPICSSHAYNHKKQNSMWAIIISVIVITIAALLILKRCFSNQRAAMEHDETLSSLFCSYDMKSF
HKTYFDLNDILEAMVDKNIVGHGGSGTVYRIELRSGDVVAVKKLWSRTRKDSTAEDQLIIKKCLNTEVET
LGNIRHKNIVKLYSYFSNFNCHLLVYDYMPNGNLWDALHKGWFHLDWPNRHQIALGVAQGLAYLHHDLLP
PIIHRDIKSTNILLDINYQPKVADFGIAKVLKDSTSTIIAGTYGYLAPEYAYSNKATTKCDVYSFGVVLM
ELITGKKPVDADFGEYKNIVYWVTTKLDTKEGVMEVIDKNLSGSFKDEMIQVLRIAMCCTCKNPSQRPTM
NEVVQLLIQTDPCLTDPYKFSTKTREASNVTENQFEEVES

ジョブの実行方法

python3 docker/run_docker.py \
  –fasta_paths=CXB51.fasta \
  –model_preset=monomer \
  –max_template_date=2024-04-01 \
  –data_dir=[データファイルのディレクトリ] \
  –output_dir=[出力結果先] \
–benchmark=true

ベンチマーク結果

ベンチマークNo.4
機器の特徴 (2) Intel Xeon Gold 6426Y 2.5GHz 16C/32T + (2) NVIDIA RTX A6000
ベンチマーク実施日2024年6月16日
実行環境詳細HPC-ProServer DPrR7960
CPU : (4) Intel Xeon Gold 6426Y 2.5GHz 16C/32T
Mem : 256GB (8) 32GB DDR5 4800MHz
SSD(sys) : 1TB NVMe SSD
SSD(data) : 15.36TB U.2 NVMe SSD <= データはこの領域に配置
GPU : (2) NVIDIA RTX A6000
OS : RockyLinux 8.10
NVIDIA Driver : 555.42.02
Docker version 26.1.3, build b72abbb
AlphaFold Version2.3.2
timeコマンド0.12user 0.01system 7:29:19elapsed 0%CPU (0avgtext+0avgdata 26700maxresident)k
13808inputs+360outputs (83major+4879minor)pagefaults 0swaps
timings.json{
“features”: 14510.03740644455,
“process_features_model_1_pred_0”: 16.09257435798645,
“predict_and_compile_model_1_pred_0”: 1369.8961718082428,
“predict_benchmark_model_1_pred_0”: 1302.927805185318,
“process_features_model_2_pred_0”: 20.470452547073364,
“predict_and_compile_model_2_pred_0”: 1240.4826309680939,
“predict_benchmark_model_2_pred_0”: 1192.4554507732391,
“process_features_model_3_pred_0”: 19.54660701751709,
“predict_and_compile_model_3_pred_0”: 1219.1449038982391,
“predict_benchmark_model_3_pred_0”: 1180.6673691272736,
“process_features_model_4_pred_0”: 19.771172761917114,
“predict_and_compile_model_4_pred_0”: 1214.8154442310333,
“predict_benchmark_model_4_pred_0”: 1179.734739780426,
“process_features_model_5_pred_0”: 21.067474603652954,
“predict_and_compile_model_5_pred_0”: 1111.5943937301636,
“predict_benchmark_model_5_pred_0”: 1075.6471076011658,
“relax_model_1_pred_0”: 153.08944416046143
}
備考features : 14510.03740644455 sec , features以外 : 12337.40374 sec
GPUメモリ:約29GB使用