{"id":1530,"date":"2019-02-07T09:47:50","date_gmt":"2019-02-07T00:47:50","guid":{"rendered":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/?p=1530"},"modified":"2021-08-02T12:37:52","modified_gmt":"2021-08-02T03:37:52","slug":"qbmg-quasi-biogenic-molecule-generator-zheng-et-al-2019-%e3%82%92%e8%a9%a6%e3%81%99","status":"publish","type":"post","link":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/2019\/02\/07\/qbmg-quasi-biogenic-molecule-generator-zheng-et-al-2019-%e3%82%92%e8%a9%a6%e3%81%99\/","title":{"rendered":"QBMG (quasi-biogenic molecule generator) (Zheng et al., 2019) \u3092\u8a66\u3059"},"content":{"rendered":"

\u7814\u7a76\u5ba4\u5185\u30e1\u30e2\uff0e \u8ad6\u6587\u7d39\u4ecb\uff0e<\/p>\n

\u74b0\u5883: macOS Sierra 10.14.2, CPU: 3.3 GHz Intel Core i5, Memory:  8 GB. Python 3.6.4, PyTorch 1.0.1, NumPy 1.14.2, RDKit 2018.03.1, KNIME Analytics Platform v3.7.1.v201901281154.<\/p>\n

\u201cQBMG: quasi-biogenic molecule generator with deep recurrent neural network\u201d<\/strong><\/p>\n

Zheng, S.; Yan, X.; Gu, Q.; Yang, Y.; Du, Y.; Lu, Y.; Xu, J.*<\/sup> J. Cheminform.<\/em> 2019<\/strong>, 11<\/em>, 5.<\/p>\n

DOI: 10.1186\/s13321-019-0328-9<\/a><\/p><\/blockquote>\n

\u8457\u8005\u3089\u306f\u30b2\u30fc\u30c8\u4ed8\u304d\u56de\u5e30\u578b\u30e6\u30cb\u30c3\u30c8\uff08gated recurrent unit: GRU\uff09\u56de\u5e30\u578b\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\uff08recurrent neural network\uff09 \u3092\u4f7f\u3063\u3066\u5929\u7136\u7269\u30e9\u30a4\u30af\u306a\u30d0\u30fc\u30c1\u30e3\u30eb\u5316\u5408\u7269\u30e9\u30a4\u30d6\u30e9\u30ea\u30fc\u306e\u751f\u6210\u3092\u884c\u3063\u305f\uff0e\u672c\u624b\u6cd5\u3067\u306fCanonical SMILES\u3092\u5165\u529b\u3068\u3059\u308b\u3053\u3068\u3067\uff0c\u7acb\u4f53\u5316\u5b66\u7684\u6027\u8cea\u3092\u53d6\u308a\u5165\u308c\u3066\u3044\u308b\uff0e\u8ee2\u79fb\u5b66\u7fd2\uff08transfer learning\uff09\u3068\u7d44\u307f\u5408\u308f\u305b\u308b\u3053\u3068\u3067\uff0c\u7279\u5b9a\u306e\u9aa8\u683c\uff08\u8ad6\u6587\u306e\u4f8b\u306f\u30af\u30de\u30ea\u30f3\uff09\u3092\u6301\u3064\u30d0\u30fc\u30c1\u30e3\u30eb\u5316\u5408\u7269\u30e9\u30a4\u30d6\u30e9\u30ea\u30fc\u306e\u751f\u6210\u3082\u53ef\u80fd\uff0e<\/p>\n

\u30b3\u30fc\u30c9<\/h2>\n

https:\/\/github.com\/SYSU-RCDD\/QBMG<\/a> \u304b\u3089\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\uff0e<\/p>\n

\u7528\u8a9e\u30e1\u30e2<\/h2>\n

SMILES<\/h3>\n

SMILES<\/span> (Simplified molecular input line entry system): \u5316\u5408\u7269\u69cb\u9020\u306e\u8868\u8a18\u6cd5\uff0e\u5316\u5408\u7269\u69cb\u9020\u3092\u5e79\u3068\u679d\u3067\u8868\u308f\u3057\uff0c\u74b0\u3092\u5dfb\u304f\u4f4d\u7f6e\u306f\u6570\u5b57\u306e\u30da\u30a2\u3067\u6307\u5b9a\u3059\u308b\uff0e<\/p>\n

\u4f8b\u3048\u3070 “CCCCCC” \u306fn<\/span>-hexane\u3067\uff0c “C1CCCCC1” \u306fcyclohexane\uff0c “C1(CO)CCCCC1” \u306fcyclohexylmethanol\u3068\u306a\u308b\uff0e<\/p>\n

\u3042\u308b\u5316\u5408\u7269\u306b\u5bfe\u3057\u3066\u8907\u6570\u306eSMILES\u6587\u5b57\u5217\u3092\u5272\u5f53\u3066\u308b\u3053\u3068\u304c\u53ef\u80fd\u3067\u3042\u308b\uff0e\u4f8b\u3048\u3070,CCO<\/code>, OCC<\/code>, \u304a\u3088\u3073C(O)C<\/code>\u306f\u3044\u305a\u308c\u3082ethanol\u306b\u5bfe\u3059\u308bSMILES\u6587\u5b57\u5217\u3068\u306a\u308b\uff0e\u4efb\u610f\u306e\u5206\u5b50\u306b\u540c\u4e00\u306eSMILES\u6587\u5b57\u5217\u3092\u751f\u6210\u3059\u308b\u30a2\u30eb\u30b4\u30ea\u30ba\u30e0\u304c\u958b\u767a\u3055\u308c\u3066\u3044\u3066\uff0c\u3053\u306e\u30a2\u30eb\u30b4\u30ea\u30ba\u30e0\u3067\u751f\u6210\u3055\u308c\u305f\u6587\u5b57\u5217\u3092 “Canonical\uff08\u6b63\u6e96\uff09 SMILES” \u3068\u547c\u3076\uff0e\u7acb\u4f53\u5316\u5b66\u60c5\u5831\u3092\u542b\u3093\u3060\u6587\u5b57\u5217\u306f “Isomeric SMILES” \u3068\u547c\u3070\u308c\u308b\uff0e<\/p>\n

\u7acb\u4f53\u5316\u5b66\u3092\u542b\u3080\u4f8b: (2R<\/span>)-2-Amino-2-phenylethanol\u306eIsomeric SMILES\u306f “C1=CC=C(C=C1)[C@H](CO)N”\uff0c(2S<\/span>)-2-Amino-2-phenylethanol\u306eIsomeric SMILES\u306f “C1=CC=C(C=C1)[C@@H](CO)N”\uff0e<\/p>\n

\u56de\u5e30\u7d50\u5408\u578b\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af<\/h3>\n

\u56de\u5e30\u7d50\u5408\u578b\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af<\/span> (recurrent neural network) \u306f\uff0c\u9806\u756a\u306b\u610f\u5473\u304c\u3042\u308b\u30c7\u30fc\u30bf\uff08\u6587\u7ae0\u3084\u6642\u7cfb\u5217\u30c7\u30fc\u30bf\u306a\u3069\uff09\u3092\u5165\u529b\u306b\u53d6\u3063\u3066\uff0c\u6587\u7ae0\u3067\u3042\u308c\u3070\u5358\u8a9e\u9593\u306e\u3064\u306a\u304c\u308a\u3092\u5b66\u7fd2\u3067\u304d\u308b\uff0e\u6a5f\u68b0\u7ffb\u8a33\u3084\u97f3\u58f0\u8a8d\u8b58\u306b\u5fdc\u7528\u3055\u308c\u3066\u6210\u529f\u3092\u53ce\u3081\u3066\u3044\u308b\uff0e\u672c\u8ad6\u6587\u306e\u5834\u5408\u306f\uff0c\u6587\u7ae0\u3067\u4f8b\u3048\u308b\u306a\u3089\u3070\uff0c\u6587\u306e\u4ee3\u308f\u308a\u306bSMILES\u3092\u5165\u529b\u3059\u308b\u306e\u3067\uff0c”C”\uff0c”1″\uff0c”=”\uff0c”[C@@H]” \u306a\u3069\u304c\u305d\u308c\u305e\u308c\u300c\u5358\u8a9e\u300d\u3068\u306a\u308b\uff0e<\/p>\n

\u8ee2\u79fb\u5b66\u7fd2<\/h3>\n

\u65e2\u5b58\u306e\u5b66\u7fd2\u6e08\u30e2\u30c7\u30eb\uff08\u51fa\u529b\u5c64\u4ee5\u5916\u306e\u90e8\u5206\uff09\u3092\u3001\u91cd\u307f\u30c7\u30fc\u30bf\u306f\u5909\u66f4\u305b\u305a\u306b\u7279\u5fb4\u91cf\u62bd\u51fa\u6a5f\u3068\u3057\u3066\u5229\u7528\u3059\u308b\u3002<\/p>\n

Tokenization<\/h3>\n

SMILES\u6587\u5b57\u5217\u3092\u300c\u5358\u8a9e = token\u300d\u306b\u5206\u89e3\u3057\u3066\uff0cToken\u30d9\u30af\u30c8\u30eb\u306b\u5909\u63db\u3059\u308b\u5de5\u7a0b\uff0e\u8ad6\u6587\u3067\u306fToken\u306e\u7a2e\u985e\u306f87\u7a2e\u985e\u306a\u306e\u3067\uff0c\u4efb\u610f\u306eToken\u306f\u9577\u305587\u306eOne-hot<\/a>\u30d9\u30af\u30c8\u30eb\uff081\u3064\u3060\u3051\u5024\u304c1\u3067\u6b8b\u308a\u304c\u5168\u30660\u306e\u30d9\u30af\u30c8\u30eb\uff09\u3068\u3057\u3066\u8868\u73fe\u3055\u308c\uff0c\u6587\u5b57\u5217\u306fOne-hot\u884c\u5217\u3068\u3057\u3066\u8868\u73fe\u3055\u308c\u308b\uff0e\u3053\u308c\u306f\u30e1\u30e2\u30ea\u3092\u5927\u5e45\u306b\u98df\u3046\u305f\u3081\uff0c\u8ad6\u6587\u3067\u306f\u3053\u306e\u884c\u5217\u3092\u81ea\u7136\u8a00\u8a9e\u51e6\u7406\u3067\u4f7f\u308f\u308c\u308bword embedding\u6cd5\u3092\u5fdc\u7528\u3057\u3066\uff0c\u60c5\u5831\u304c\u8c4a\u304b\u306aword embedding\u884c\u5217\u3078\u3068\u5909\u63db\u3057\u3066\u304b\u3089\uff0c\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306b\u5165\u529b\u3057\u3066\u3044\u308b\uff0e<\/p>\n

\u30b3\u30fc\u30c9\u306e\u4fee\u6b63<\/h2>\n

CPU\u4e0a\u3067\u52d5\u304b\u3059\u306e\u3068\uff0cPyTorch\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u304c\u9055\u3046\u306e\u304c\u7406\u7531\u3067\u30a8\u30e9\u30fc\u304c\u51fa\u308b\u305f\u3081\uff0c\u30a8\u30e9\u30fc\u306b\u3057\u305f\u304c\u3063\u3066\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u4fee\u6b63\u3057\u307e\u3057\u305f\uff0e<\/p>\n

transfer-learning.py<\/code>\u306e25\u884c\u76ee\u3068sample.py<\/code>\u306e18\u884c\u76ee<\/p>\n

\nPrior.rnn.load_state_dict(torch.load(restore_from))\n<\/pre>\n
\u3092<\/p>\n
\nPrior.rnn.load_state_dict(torch.load(restore_from, map_location=cpu))\n<\/pre>\n
\u306b\u4fee\u6b63\uff0etransfer-learning.py<\/code>\u306e49\u884c\u76ee<\/p>\n
        tqdm.write("Epoch {:3d}   step {:3d}    loss: {:5.2f}\\n".format(epoch, step, loss.data[0]))\n<\/pre>\n
\u3092<\/p>\n
        tqdm.write("Epoch {:3d}   step {:3d}    loss: {:5.2f}\\n".format(epoch, step, loss.item()))\n<\/pre>\n
\u306b\u4fee\u6b63\uff0e<\/p>\n
Quasi-biogenic molecules\u306e\u751f\u6210<\/h2>\n
\u8ad6\u6587\u3067\u306f\u7121\u6599\u306e\u5316\u5408\u7269\u69cb\u9020\u30c7\u30fc\u30bf\u30d9\u30fc\u30b9ZINC<\/a>\u306ebiogenic compounds 153,733\u69cb\u9020\u3092\u4f7f\u3063\u3066RNN\u3092\u8a13\u7df4\u3057\u3066\uff0c\u8ee2\u79fb\u5b66\u7fd2\u3067\u306f\u30af\u30de\u30ea\u30f3\u9aa8\u683c\u3092\u6301\u3064\u5316\u5408\u72692239\u7a2e\u985e\u3092\u4f7f\u3063\u3066\u3044\u307e\u3057\u305f\uff0e \u4eca\u56de\u306fphorbol esters\u3068daphnanes\u3068ingenol esters\uff08\u3068\u305d\u308c\u3089\u306e\u30a2\u30ca\u30ed\u30b0\uff09\u5408\u308f\u305b\u3066212\u5316\u5408\u7269 (test.smi\u3068\u3057\u3066data\u30d5\u30a9\u30eb\u30c0\u306b\u4fdd\u5b58) \u3092\u4f7f\u3063\u3066\u8ee2\u79fb\u5b66\u7fd2\u3068\u69cb\u9020\u306e\u751f\u6210\u3092\u884c\u3063\u3066\u307f\u307e\u3057\u305f\uff0e<\/p>\n
\"\"<\/p>\n
(>anaconda3-5.1.0)> $ python transfer_learning.py .\/data\/test.smi\n<\/code><\/pre>\n
100 epochs\u306e\u5b66\u7fd2\u306f17\u520630\u79d2\u3067\u7d42\u4e86\u3057\u307e\u3057\u305f\uff0eValid SMILES\u306e\u5272\u5408\u306f\uff0c10 epochs\u306794%\uff0c\u5f8c\u306f96.0\u2013100.0%\u306e\u9593\u3067\u3057\u305f\uff0e<\/p>\n
\u6b21\u306b\u4fdd\u5b58\u3055\u308c\u305f .\/data\/100_epochs_transfer.ckpt \u3092\u4f7f\u3063\u3066\u69cb\u9020\u306e\u751f\u6210\u3092\u884c\u3044\u307e\u3057\u305f\uff08test.ckpt\u3068\u3057\u3066\u30b3\u30d4\u30fc\u3057\u3066\u304a\u304f\uff09\uff0e \u30b3\u30de\u30f3\u30c9\u306e\u6700\u5f8c\u306e\u6570\u5b57\u304c\u751f\u6210\u3059\u308b\u69cb\u9020\u306e\u6570\u3067\u3059\uff0e<\/p>\n
(>anaconda3-5.1.0)> $ python sample.py .\/data\/test.ckpt 1000\n<\/code><\/pre>\n
4\u520637\u79d2\u3067\u7d42\u4e86\u3057\uff0csample.smi\u30d5\u30a1\u30a4\u30eb\u304c\u751f\u6210\u3055\u308c\u307e\u3057\u305f\uff0e<\/p>\n
\u69cb\u9020\u306e\u78ba\u8a8d<\/h2>\n
KNIME\u3067\u5143\u306etest.smi\u3068\u91cd\u8907\u3057\u305f\u69cb\u9020\uff08210\u5316\u5408\u7269\uff09\u3092\u9664\u304f\u3068\uff0c791\u500b\u306e\u65b0\u898f\u69cb\u9020\u304c\u751f\u6210\u3055\u308c\u3066\u3044\u307e\u3057\u305f\uff0e\u3068\u308a\u3042\u3048\u305a\u5316\u5b66\u69cb\u9020\u5f0f\u3092PDF\u3067\u51fa\u529b\u3057\u3066\u76ee\u3067\u78ba\u8a8d\uff0e
\n\"\"<\/p>\n
\u8ee2\u79fb\u5b66\u7fd2\u306b\u4f7f\u7528\u3057\u305f\u30b8\u30c6\u30eb\u30da\u30ce\u30a4\u30c9\u306e\u69cb\u9020\u3092\u305d\u306e\u307e\u307e\u53cd\u6620\u3057\u305f\u69cb\u9020\u3068\u305d\u308c\u3068\u306f\u95a2\u4fc2\u306e\u306a\u3044\u5143\u306e\u5b66\u7fd2\u6e08\u30c7\u30fc\u30bf\u7531\u6765\u3089\u3057\u3044\u69cb\u9020\u304c\u307b\u3068\u3093\u3069\u3067\u3057\u305f\u304c\uff0c\u30b8\u30c6\u30eb\u30da\u30ce\u30a4\u30c9\u306e\u69cb\u9020\u304c\u5c11\u3057\u5909\u5316\u3057\u305f\u308a\u878d\u5408\u3057\u305f\u308a\u3057\u305f\u9762\u767d\u3044\u69cb\u9020\u3082\u3044\u304f\u3064\u304b\u751f\u6210\u3055\u308c\u3066\u3044\u307e\u3057\u305f\uff0e<\/p>\n
\"\"<\/p>\n
\u65e2\u77e5\u306e\u751f\u7406\u6d3b\u6027\u5929\u7136\u7269\u306e\u30d5\u30a1\u30fc\u30de\u30b3\u30d5\u30a9\u30a2\u3092\u5225\u306e\u9aa8\u683c\u306b\u79fb\u3057\u63db\u3048\u308b\u3088\u3046\u306a\u3053\u3068\u304c\u53ef\u80fd\u306a\u3089\uff0c\u9762\u767d\u3044\u30d0\u30fc\u30c1\u30e3\u30eb\u30e9\u30a4\u30d6\u30e9\u30ea\u306b\u306a\u308a\u305d\u3046\u3067\u3059\u304c\uff0c\u4eca\u306e\u3068\u3053\u308d\u305d\u3046\u3044\u3063\u305f\u3053\u3068\u306f\u51fa\u6765\u7121\u3055\u305d\u3046\u3067\u3059\uff0e<\/p>\n
\u6b21\u306b\u8aad\u3080\u8ad6\u6587<\/h2>\n
\u201cAutomatic Chemical Design Using a Data-Driven Continuous Representation of Molecules\u201d<\/strong> G\u00f3mez-Bombarelli, R., et al<\/span>., ACS Cent. Sci.<\/em> 2018<\/strong>, 4<\/em>, 268\u2013276. DOI: 10.1021\/acscentsci.7b00572<\/a><\/p>\n
(\u4e86)<\/p>\n","protected":false},"excerpt":{"rendered":"
\u7814\u7a76\u5ba4\u5185\u30e1\u30e2\uff0e \u8ad6\u6587\u7d39\u4ecb\uff0e \u74b0\u5883: macOS Sierra …<\/p>\n","protected":false},"author":1,"featured_media":1563,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_uag_custom_page_level_css":"","_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"_uf_show_specific_survey":0,"_uf_disable_surveys":false,"_locale":"","_original_post":"","footnotes":""},"categories":[5],"tags":[],"class_list":["post-1530","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-5","ja"],"aioseo_notices":[],"uagb_featured_image_src":{"full":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol.png",1866,600,false],"thumbnail":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol-150x150.png",150,150,true],"medium":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol-300x96.png",300,96,true],"medium_large":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol-768x247.png",768,247,true],"large":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol-1024x329.png",800,257,true],"1536x1536":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol.png",1536,494,false],"2048x2048":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol.png",1866,600,false],"onepress-blog-small":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol.png",300,96,false],"onepress-small":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol.png",480,154,false],"onepress-medium":["https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp\/wp-content\/uploads\/2019\/02\/phorboldaphneningenol.png",640,206,false]},"uagb_author_info":{"display_name":"RCY","author_link":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/author\/charlesy\/"},"uagb_comment_info":0,"uagb_excerpt":"\u7814\u7a76\u5ba4\u5185\u30e1\u30e2\uff0e \u8ad6\u6587\u7d39\u4ecb\uff0e \u74b0\u5883: macOS Sierra ...","_links":{"self":[{"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/posts\/1530","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/comments?post=1530"}],"version-history":[{"count":70,"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/posts\/1530\/revisions"}],"predecessor-version":[{"id":2798,"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/posts\/1530\/revisions\/2798"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/media\/1563"}],"wp:attachment":[{"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/media?parent=1530"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/categories?post=1530"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ag.kagawa-u.ac.jp\/charlesy\/wp-json\/wp\/v2\/tags?post=1530"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}