{"id":1077171,"date":"2024-01-31T12:07:00","date_gmt":"2024-01-31T20:07:00","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-blog-post&#038;p=1077171"},"modified":"2024-09-25T04:02:12","modified_gmt":"2024-09-25T11:02:12","slug":"visnet","status":"publish","type":"msr-blog-post","link":"https:\/\/www.microsoft.com\/en-us\/research\/articles\/visnet\/","title":{"rendered":"ViSNet\uff1a\u7528\u4e8e\u5206\u5b50\u6027\u8d28\u9884\u6d4b\u548c\u52a8\u529b\u5b66\u6a21\u62df\u7684\u901a\u7528\u5206\u5b50\u7ed3\u6784\u5efa\u6a21\u7f51\u7edc"},"content":{"rendered":"\n<p>\u7f16\u8005\u6309\uff1a\u5c3d\u7ba1\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\u5df2\u7ecf\u5f7b\u5e95\u98a0\u8986\u4e86\u5206\u5b50\u5efa\u6a21\u9886\u57df\uff0c\u4f46\u6700\u5148\u8fdb\u7684\u7b97\u6cd5\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u4ecd\u7136\u9762\u4e34\u7740\u51e0\u4f55\u4fe1\u606f\u5229\u7528\u4e0d\u8db3\u548c\u9ad8\u6602\u8ba1\u7b97\u6210\u672c\u7684\u963b\u788d\u3002\u4e3a\u6b64\uff0c\u5fae\u8f6f\u7814\u7a76\u9662\u79d1\u5b66\u667a\u80fd\u4e2d\u5fc3\uff08Microsoft Research AI4Science\uff09\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u901a\u7528\u5206\u5b50\u7ed3\u6784\u5efa\u6a21\u7f51\u7edc ViSNet\u3002\u5728\u591a\u4e2a\u5206\u5b50\u52a8\u529b\u5b66\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cViSNet \u5747\u8868\u73b0\u4f18\u5f02\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>\u5206\u5b50\u51e0\u4f55\u5efa\u6a21\u5728\u7406\u89e3\u751f\u7269\u6d3b\u6027\u673a\u5236\u3001\u5316\u5b66\u6027\u8d28\u9884\u6d4b\u3001\u836f\u7269\u8bbe\u8ba1\u548c\u86cb\u767d\u8d28\u5de5\u7a0b\u65b9\u9762\u53d1\u6325\u7740\u5173\u952e\u4f5c\u7528\u3002\u7136\u800c\uff0c\u867d\u7136\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\uff08geometric deep learning\uff09\u662f\u4e00\u79cd\u4f4e\u6210\u672c\u3001\u9ad8\u7cbe\u5ea6\u4e14\u53ef\u4ee5\u88ab\u5e7f\u6cdb\u4f7f\u7528\u7684\u8ba1\u7b97\u65b9\u6cd5\uff0c\u5728\u8fc7\u53bb\u5341\u5e74\u53d6\u5f97\u4e86\u5de8\u5927\u8fdb\u5c55\uff0c\u4f46\u8fd9\u79cd\u6280\u672f\u4ecd\u7136\u5b58\u5728\u4e00\u4e9b\u6709\u5f85\u89e3\u51b3\u7684\u95ee\u9898\u548c\u5c40\u9650\u6027\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u5206\u5b50\u53ef\u89e3\u91ca\u6027\u4e0d\u8db3\uff1a<\/strong>\u6df1\u5c42\u795e\u7ecf\u7f51\u7edc\u5c3d\u7ba1\u53ef\u4ee5\u8fdb\u884c\u9884\u6d4b\uff0c\u4f46\u7f3a\u4e4f\u5bf9\u5206\u5b50\u7684\u6df1\u5165\u6d1e\u5bdf\uff1b<\/li>\n\n\n\n<li><strong>\u968f\u7740\u5206\u5b50\u5c3a\u5bf8\u7684\u589e\u52a0\uff0c\u8ba1\u7b97\u6210\u672c\u8fc5\u901f\u589e\u52a0\uff1a<\/strong>\u4e00\u4e9b\u76ee\u524d\u6700\u5148\u8fdb\u7684\u65b9\u6cd5\u4e2d\u91c7\u7528\u7684\u9ad8\u9636 Clebsch-Gordan \u7cfb\u6570\u8ba1\u7b97\u662f\u8ba1\u7b97\u5bc6\u96c6\u578b\u7684\uff0c\u56e0\u6b64\u963b\u788d\u4e86\u5176\u5728\u5927\u5206\u5b50\u4e2d\u7684\u5e94\u7528\uff1b<\/li>\n\n\n\n<li><strong>\u9700\u8981\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u76f2\u76ee\u6d4b\u8bd5\u548c\u8bc4\u4f30\uff1a<\/strong>\u6a21\u578b\u603b\u662f\u5728\u57fa\u51c6\u6d4b\u8bd5\u4e0a\u8fdb\u884c\u6d4b\u8bd5\uff0c\u540c\u65f6\u4e5f\u9700\u8981\u4ed4\u7ec6\u8bc4\u4f30\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u6709\u6548\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u96be\u9898\uff0c\u5fae\u8f6f\u7814\u7a76\u9662\u79d1\u5b66\u667a\u80fd\u4e2d\u5fc3\u7684\u7814\u7a76\u5458\u4eec\u5c06\u7814\u7a76\u91cd\u70b9\u805a\u7126\u5728\u4e86\u5982\u4f55\u63d0\u9ad8\u5206\u5b50\u53ef\u89e3\u91ca\u6027\u3001\u964d\u4f4e\u8ba1\u7b97\u6210\u672c\u4ee5\u53ca\u8bc4\u4f30\u5b9e\u9645\u5e94\u7528\u51e0\u4e2a\u65b9\u9762\uff0c\u5e76\u521b\u65b0\u6027\u5730\u63d0\u51fa\u4e86\u901a\u7528\u5206\u5b50\u7ed3\u6784\u5efa\u6a21\u7f51\u7edc ViSNet (Vector-Scalar interactive graph neural Network)\u3002\u76f8\u5173\u6587\u7ae0\u201cEnhancing geometric representations for molecules with equivariant vector-scalar interactive message passing\u201d<strong>\u5df2\u53d1\u8868\u5728\u300a\u81ea\u7136-\u901a\u8baf\u300b\uff08Nature Communications\uff09\u6742\u5fd7\u4e0a\uff0c\u5e76\u540c\u65f6\u5165\u9009\u4e86\u201cAI and machine learning\u201d\u548c\u201cBiotechnology and method\u201d\u4e24\u4e2a\u9886\u57df\u7684\u7f16\u8f91\u7cbe\u9009\u6587\u7ae0\u3002<\/strong><\/p>\n\n\n\n<p>ViSNet\u8bba\u6587\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.nature.com\/articles\/s41467-023-43720-2\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.nature.com\/articles\/s41467-023-43720-2<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u201cAI and machine learning\u201d\u9886\u57df\u7f16\u8f91\u7cbe\u9009\u6587\u7ae0\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.nature.com\/collections\/ceiajcdbeb\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.nature.com\/collections\/ceiajcdbeb<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u201cBiotechnology and method\u201d\u9886\u57df\u7f16\u8f91\u7cbe\u9009\u6587\u7ae0\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.nature.com\/collections\/idhhgedgig\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.nature.com\/collections\/idhhgedgig<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u6709\u6548\u63d0\u5347\u5206\u5b50\u51e0\u4f55\u8868\u793a\">\u6709\u6548\u63d0\u5347\u5206\u5b50\u51e0\u4f55\u8868\u793a<\/h2>\n\n\n\n<p>\u7814\u7a76\u5458\u4eec\u6700\u521d\u8ba1\u5212\u901a\u8fc7\u6709\u6548\u4e14\u5145\u5206\u5730\u5229\u7528\u5206\u5b50\u7ed3\u6784\u7684\u9886\u57df\u77e5\u8bc6\u6765\u8bbe\u8ba1\u6a21\u578b\u3002\u7531\u4e8e\u7ecf\u5178\u5206\u5b50\u52a8\u529b\u5b66\uff08molecular dynamics, MD\uff09\u901a\u8fc7\u660e\u786e\u63cf\u8ff0\u52bf\u80fd\u51fd\u6570\u4e2d\u7684\u952e\u957f\u3001\u952e\u89d2\u548c\u4e8c\u9762\u89d2\u6765\u6a21\u62df\u5206\u5b50\u8fd0\u52a8\uff0c\u6240\u4ee5\u53d7\u7ecf\u5178 MD \u6a21\u62df\u7684\u542f\u53d1\uff0c\u7814\u7a76\u5458\u4eec\u5c06\u8fd9\u4e9b\u9879\u76ee\u8f6c\u6362\u5e76\u6269\u5c55\uff0c\u4ece\u800c\u6784\u5efa\u4e86 ViSNet \u72ec\u7279\u7684\u6a21\u578b\u8bbe\u8ba1\u3002<\/p>\n\n\n\n<p>\u4e0e\u901a\u8fc7\u7b80\u5355\u7684\u7279\u5f81\u5de5\u7a0b\u8fc7\u7a0b\u76f4\u63a5\u91c7\u7528\u89d2\u5ea6\u6216\u4e8c\u9762\u4f53\u4fe1\u606f\u4e0d\u540c\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u201c\u65b9\u5411\u5355\u5143\u201d\u8fd9\u4e2a\u6982\u5ff5\u4f5c\u4e3a\u8282\u70b9\u7684\u5411\u91cf\u5316\u8868\u793a\uff0c\u5373\u4ece\u4e2d\u5fc3\u8282\u70b9\u5230\u5176\u4efb\u4f55\u7b2c\u4e00\u4e2a\u76f8\u90bb\u8282\u70b9\u7684\u6240\u6709\u5f52\u4e00\u5316\u5411\u91cf\u7684\u603b\u548c\uff0c\u4f5c\u4e3a\u4e2d\u5fc3\u8282\u70b9\u7684\u77e2\u91cf\u5316\u8868\u793a\u3002\u518d\u4ee5\u6b64\u5c06\u952e\u957f\u3001\u952e\u89d2\u548c\u4e8c\u9762\u89d2\u8ba1\u7b97\u6269\u5c55\u5230\u4e8c\u4f53\u3001\u4e09\u4f53\u548c\u56db\u4f53\u76f8\u4e92\u4f5c\u7528\u3002\u7136\u540e\uff0c\u901a\u8fc7\u8bbe\u8ba1\u8fd0\u884c\u65f6\u51e0\u4f55\u8ba1\u7b97\uff08runtime geometry calculation, RGC\uff09\u6a21\u5757\u6765\u63cf\u8ff0\u6a21\u578b\u64cd\u4f5c\u7b49\u591a\u4f53\u4ea4\u4e92\u3002<\/p>\n\n\n\n<p>\u66f4\u91cd\u8981\u7684\u662f\uff0c\u4e09\u4f53\u548c\u56db\u4f53\u76f8\u4e92\u4f5c\u7528\u7684 RGC \u8ba1\u7b97\u90fd\u53ea\u6709\u7ebf\u6027\u65f6\u95f4\u590d\u6742\u5ea6\u3002\u56e0\u6b64\uff0c\u7814\u7a76\u5458\u4eec\u53c8\u8fdb\u4e00\u6b65\u63d0\u51fa\u4e86\u5411\u91cf\u6807\u91cf\u4ea4\u4e92\u5f0f\u6d88\u606f\u4f20\u9012\u673a\u5236\uff08ViS-MP\uff09\uff0c\u5176\u4e2d\u65b9\u5411\u5355\u5143\u4f1a\u901a\u8fc7\u6784\u5efa\u5757\u7531\u8282\u70b9\u548c\u8fb9\u7684\u6807\u91cf\u8868\u793a\u8fed\u4ee3\u66f4\u65b0\uff0c\u53cd\u8fc7\u6765\uff0c\u6807\u91cf\u8868\u793a\u7531\u65b9\u5411\u5355\u5143\u540c\u65f6\u66f4\u65b0 RGC \u6a21\u5757\u3002RGC \u548c ViS-MP \u7684\u72ec\u7279\u8bbe\u8ba1\u663e\u8457\u589e\u5f3a\u4e86\u51e0\u4f55\u7f16\u7801\u80fd\u529b\u5e76\u52a0\u901f\u4e86\u5206\u5b50\u56fe\u795e\u7ecf\u7f51\u7edc\u4e2d\u7684\u6d88\u606f\u4f20\u9012\u8fc7\u7a0b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1430\" height=\"826\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig1.png\" alt=\"VisNet diagram\" class=\"wp-image-1002630\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig1.png 1430w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig1-300x173.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig1-1024x591.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig1-768x444.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig1-240x139.png 240w\" sizes=\"auto, (max-width: 1430px) 100vw, 1430px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe1\uff1aViSNet \u7f51\u7edc\u7ed3\u6784\u793a\u610f\u56fe<\/em><\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"visnet\u5728\u5206\u5b50\u5efa\u6a21\u548c\u6027\u8d28\u9884\u6d4b\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u8868\u73b0\">ViSNet\u5728\u5206\u5b50\u5efa\u6a21\u548c\u6027\u8d28\u9884\u6d4b\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u8868\u73b0<\/h2>\n\n\n\n<p>\u7814\u7a76\u5458\u4eec\u9996\u5148\u5c06 ViSNet \u5728\u5e7f\u6cdb\u4f7f\u7528\u7684\u5206\u5b50\u5316\u5b66\u6027\u8d28\u9884\u6d4b\u57fa\u51c6\u4e0a\u8fdb\u884c\u4e86\u8bc4\u4f30\u3002\u5728 MD17\u3001\u4fee\u8ba2\u7248 MD17 \u3001 MD22\u3001QM9 \u4ee5\u53ca Molecule3D \u6570\u636e\u96c6\u4e0a\u663e\u793a\u51fa\u5353\u8d8a\u7684\u6027\u80fd\uff0c\u8bc1\u660e\u4e86\u5206\u5b50\u51e0\u4f55\u8868\u793a\u7684\u5f3a\u5927\u80fd\u529b\u3002\u7136\u540e\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u5728\u81ea\u5df2\u5f00\u53d1\u7684 DFT\uff08\u5bc6\u5ea6\u51fd\u6570\u7406\u8bba\uff09\u7cbe\u5ea6\u7684\u86cb\u767d\u8d28\u6570\u636e\u96c6 AIMD-Chig \u6570\u636e\u96c6\u4e0a\u8bad\u7ec3\u4e86 ViSNet\uff0c\u5e76\u5bf9\u86cb\u767d\u8d28 Chignolin \u8fdb\u884c\u4e86 MD \u6a21\u62df\u3002<\/p>\n\n\n\n<p>ViSNet \u53d6\u5f97\u4e86\u6bd4\u7ecf\u9a8c\u529b\u573a\u548c\u73b0\u4ee3\u673a\u5668\u5b66\u4e60\u529b\u573a\u66f4\u597d\u7684\u6027\u80fd\u53ca\u4ee4\u4eba\u6ee1\u610f\u7684\u7ed3\u679c\u3002ViSNet \u7684\u6a21\u62df\u7ed3\u679c\u4e0e\u5728 DFT \u6c34\u5e73\u4e0a\u83b7\u5f97\u7684\u7ed3\u679c\u975e\u5e38\u63a5\u8fd1\uff0c\u8fd9\u8868\u660e ViSNet \u5728\u6570\u636e\u6548\u7387\u548c\u6a21\u62df\u4fdd\u771f\u5ea6\u65b9\u9762\u5177\u6709\u6f5c\u529b\u3002<\/p>\n\n\n\n<p>\u7814\u7a76\u5458\u4eec\u7528 ViSNet \u53c2\u52a0\u4e86\u5168\u7403\u9996\u5c4a AI \u836f\u7269\u7814\u53d1\u7b97\u6cd5\u5927\u8d5b\u3002\u8be5\u5927\u8d5b\u65e8\u5728\u6839\u636e\u5c0f\u5206\u5b50\u7684\u5e8f\u5217\u4fe1\u606f\uff08\u5373SMILES\uff09\u9884\u6d4b\u9488\u5bf9\u65b0\u51a0\u75c5\u6bd2 SARS-CoV-2 \u4e3b\u8981\u86cb\u767d\u9176\u7684\u6291\u5236\u5242\u3002\u5171\u6709\u6765\u81ea\u5168\u7403878\u652f\u56e2\u961f\u76841105\u540d\u53c2\u8d5b\u8005\u53c2\u4e0e\u4e86\u6b64\u6b21\u6bd4\u8d5b\u3002\u6700\u7ec8\uff0c\u7814\u7a76\u5458\u4eec\u51ed\u501f ViSNet \u83b7\u5f97\u4e86\u6bd4\u8d5b\u7684\u603b\u51a0\u519b\uff0c\u4e5f\u5c55\u73b0\u4e86 ViSNet \u4f18\u5f02\u7684\u9884\u6d4b\u51c6\u786e\u6027\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u5982\u4f55\u83b7\u53d6visnet\u6a21\u578b\">\u5982\u4f55\u83b7\u53d6ViSNet\u6a21\u578b\uff1f<\/h2>\n\n\n\n<p>\u4e3a\u4e86\u4fc3\u8fdb\u66f4\u5e7f\u6cdb\u7684\u5e94\u7528\u548c\u4fbf\u6377\u7684\u4f7f\u7528\uff0cViSNet \u5df2\u88ab\u5fae\u8f6f\u7eb3\u5165 Pytorch Geometry \u5e93\uff0c\u4f5c\u4e3a\u5206\u5b50\u5efa\u6a21\u548c\u5c5e\u6027\u9884\u6d4b\u9886\u57df\u7684\u57fa\u672c\u6a21\u578b\u3002ViSNet \u7684\u5b9a\u671f\u7ef4\u62a4\u548c\u66f4\u65b0\u7248\u672c\u4e5f\u53ef\u5728 GitHub\u4e0a \u83b7\u53d6\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1430\" height=\"728\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig2.png\" alt=\"models.VisNet screenshot\" class=\"wp-image-1002633\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig2.png 1430w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig2-300x153.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig2-1024x521.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig2-768x391.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig2-240x122.png 240w\" sizes=\"auto, (max-width: 1430px) 100vw, 1430px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe2\uff1aViSNet \u5728 Pytorch Geometry \u4e2d\u4f5c\u4e3a\u57fa\u7840\u6a21\u578b<\/em><\/figcaption><\/figure>\n\n\n\n<p>Pytorch Geometry \u5e93\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/pytorch-geometric.readthedocs.io\/en\/latest\/generated\/torch_geometric.nn.models.ViSNet.html\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/pytorch-geometric.readthedocs.io\/en\/latest\/generated\/torch_geometric.nn.models.ViSNet.html<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>GitHub \u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/github.com\/microsoft\/AI2BMD\/tree\/ViSNet\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/microsoft\/AI2BMD\/tree\/ViSNet<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u6b64\u5916\uff0c\u8003\u8651\u5230\u56fe\u795e\u7ecf\u7f51\u7edc\u968f\u7740\u6a21\u578b\u53d8\u5f97\u8d8a\u6765\u8d8a\u5927\u3001\u8d8a\u6765\u8d8a\u6df1\uff0c\u53ef\u80fd\u4f1a\u9047\u5230\u201c\u8fc7\u5ea6\u5e73\u6ed1\u201d\u7684\u98ce\u9669\uff0c\u7814\u7a76\u5458\u4eec\u8fd8\u8fdb\u4e00\u6b65\u8bbe\u8ba1\u4e86 ViSNet \u7684 Transformer \u7248\u672c\uff0c\u53ef\u4ee5\u5c06 RGC \u6a21\u5757\u8f6c\u79fb\u5230 Transformer \u6ce8\u610f\u529b\u8ba1\u7b97\u4e2d\uff0c<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.msra.cn\/zh-cn\/news\/features\/new-arrival-in-research-5#research3\">\u5e76\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u9896\u7684\u539f\u5b50\u95f4\u4f4d\u7f6e\u7f16\u7801\uff08IPE\uff09\uff0c\u547d\u540d\u4e3a Geoformer\uff08Geometric Transformer\u7684\u7f29\u5199\uff09\u3002\u4f5c\u4e3a ViSNet \u7684 Transformer \u53d8\u4f53\uff0cGeoformer \u53ef\u901a\u8fc7\u5806\u53e0\u6570\u767e\u4e2a\u6ce8\u610f\u529b\u5757\u6765\u8fdb\u884c\u5927\u6a21\u578b\u8bad\u7ec3\u3002\u76f8\u5173\u7814\u7a76\u8bba\u6587\u53d1\u8868\u4e8e NeuraIPS 2023\u3002<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u5206\u5b50\u52a8\u529b\u5b66\u6a21\u62df\u7684\u672a\u6765-\u517c\u5177\u4eba\u5de5\u667a\u80fd\u4e0e\u4ece\u5934\u8ba1\u7b97\u7cbe\u5ea6\u7684\u80fd\u529b\">\u5206\u5b50\u52a8\u529b\u5b66\u6a21\u62df\u7684\u672a\u6765\uff1a\u517c\u5177\u4eba\u5de5\u667a\u80fd\u4e0e\u4ece\u5934\u8ba1\u7b97\u7cbe\u5ea6\u7684\u80fd\u529b<\/h2>\n\n\n\n<p>\u4f5c\u4e3a\u4eba\u5de5\u667a\u80fd\uff08AI\uff09\u9a71\u52a8\u7684\u4ece\u5934\u7b97\u5206\u5b50\u52a8\u529b\u5b66\uff08AI<sup>2<\/sup>BMD\uff09\u9879\u76ee\u7684\u91cd\u8981\u7ec4\u6210\u90e8\u5206\uff0cViSNet \u81f4\u529b\u4e8e\u5b9e\u73b0\u52a0\u901f\u5206\u5b50\u52a8\u529b\u5b66\u6a21\u62df\u7684\u76ee\u6807\uff0c\u4f7f\u5927\u578b\u5206\u5b50\u7cfb\u7edf\u7684\u6a21\u62df\u7cbe\u5ea6\u63a5\u8fd1\u4ece\u5934\u7b97\u6cd5\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1416\" height=\"731\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig3.png\" alt=\"VisNet diagram\" class=\"wp-image-1002636\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig3.png 1416w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig3-300x155.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig3-1024x529.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig3-768x396.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/uploads\/prod\/2024\/01\/visnet_Fig3-240x124.png 240w\" sizes=\"auto, (max-width: 1416px) 100vw, 1416px\" \/><figcaption class=\"wp-element-caption\"><em>\u56fe3\uff1aAI2BMD \u6574\u4f53\u6d41\u7a0b<\/em><\/figcaption><\/figure>\n\n\n\n<p>ViSNet \u53ef\u4ee5\u8ba9 AI<sup>2<\/sup>BMD \u5b9e\u73b0\u5bf9\u5305\u542b\u8d85\u8fc710,000\u4e2a\u539f\u5b50\u7684\u86cb\u767d\u8d28\u7684\u80fd\u91cf\u548c\u529b\u8ba1\u7b97\u8fbe\u5230\u63a5\u8fd1\u4ece\u5934\u7b97\u6cd5\u7684\u7cbe\u5ea6\u3002\u5229\u7528 ViSNet \u8fdb\u884c\u86cb\u767d\u8d28\u52a8\u529b\u5b66\u6a21\u62df\u8fd8\u53ef\u63d0\u9ad8\u81ea\u7531\u80fd\u4f30\u8ba1\u7684\u51c6\u786e\u6027\uff0c\u63d0\u4f9b\u6709\u5173\u86cb\u767d\u8d28\u6298\u53e0\u70ed\u529b\u5b66\u7684\u6df1\u5165\u9884\u6d4b\uff0c\u5e76\u6709\u52a9\u4e8e\u8868\u5f81\u86cb\u767d\u8d28\u7684\u7279\u6027\uff0c\u4ece\u800c\u6f5c\u5728\u5730\u589e\u5f3a\u5b9e\u9a8c\u7814\u7a76\u3002<\/p>\n\n\n\n<p>\u76f8\u5173\u94fe\u63a5\uff1a<\/p>\n\n\n\n<p>ViSNet\u8bba\u6587\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.nature.com\/articles\/s41467-023-43720-2\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.nature.com\/articles\/s41467-023-43720-2<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>AIMD-Chig \u6570\u636e\u96c6\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.nature.com\/articles\/s41597-023-02465-9\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/www.nature.com\/articles\/s41597-023-02465-9<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u9996\u5c4aAI\u836f\u7269\u7814\u53d1\u7b97\u6cd5\u5927\u8d5b\u5b98\u65b9\u7f51\u9875\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/aistudio.baidu.com\/competition\/detail\/1012\/0\/leaderboard\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/aistudio.baidu.com\/competition\/detail\/1012\/0\/leaderboard<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>ViSNet-Pytorch Geometry \u5e93\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/pytorch-geometric.readthedocs.io\/en\/latest\/generated\/torch_geometric.nn.models.ViSNet.html\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/pytorch-geometric.readthedocs.io\/en\/latest\/generated\/torch_geometric.nn.models.ViSNet.html<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>ViSNet-GitHub\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/github.com\/microsoft\/AI2BMD\/tree\/ViSNet\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/microsoft\/AI2BMD\/tree\/ViSNet<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>Geoformer\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/github.com\/microsoft\/AI2BMD\/blob\/Geoformer\/Geoformer.pdf\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/github.com\/microsoft\/AI2BMD\/blob\/Geoformer\/Geoformer.pdf<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>AI<sup>2<\/sup>BMD\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/microsoft.github.io\/AI2BMD\/index.html\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/microsoft.github.io\/AI2BMD\/index.html<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u7f16\u8005\u6309\uff1a\u5c3d\u7ba1\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\u5df2\u7ecf\u5f7b\u5e95\u98a0\u8986\u4e86\u5206\u5b50\u5efa\u6a21\u9886\u57df\uff0c\u4f46\u6700\u5148\u8fdb\u7684\u7b97\u6cd5\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u4ecd\u7136\u9762\u4e34\u7740\u51e0\u4f55\u4fe1\u606f\u5229\u7528\u4e0d\u8db3\u548c\u9ad8\u6602\u8ba1\u7b97\u6210\u672c\u7684\u963b\u788d\u3002\u4e3a\u6b64\uff0c\u5fae\u8f6f\u7814\u7a76\u9662\u79d1\u5b66\u667a\u80fd\u4e2d\u5fc3\uff08Microsoft Research AI4Science\uff09\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u901a\u7528\u5206\u5b50\u7ed3\u6784\u5efa\u6a21\u7f51\u7edc ViSNet\u3002\u5728\u591a\u4e2a\u5206\u5b50\u52a8\u529b\u5b66\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cViSNet \u5747\u8868\u73b0\u4f18\u5f02\u3002 \u5206\u5b50\u51e0\u4f55\u5efa\u6a21\u5728\u7406\u89e3\u751f\u7269\u6d3b\u6027\u673a\u5236\u3001\u5316\u5b66\u6027\u8d28\u9884\u6d4b\u3001\u836f\u7269\u8bbe\u8ba1\u548c\u86cb\u767d\u8d28\u5de5\u7a0b\u65b9\u9762\u53d1\u6325\u7740\u5173\u952e\u4f5c\u7528\u3002\u7136\u800c\uff0c\u867d\u7136\u51e0\u4f55\u6df1\u5ea6\u5b66\u4e60\uff08geometric deep learning\uff09\u662f\u4e00\u79cd\u4f4e\u6210\u672c\u3001\u9ad8\u7cbe\u5ea6\u4e14\u53ef\u4ee5\u88ab\u5e7f\u6cdb\u4f7f\u7528\u7684\u8ba1\u7b97\u65b9\u6cd5\uff0c\u5728\u8fc7\u53bb\u5341\u5e74\u53d6\u5f97\u4e86\u5de8\u5927\u8fdb\u5c55\uff0c\u4f46\u8fd9\u79cd\u6280\u672f\u4ecd\u7136\u5b58\u5728\u4e00\u4e9b\u6709\u5f85\u89e3\u51b3\u7684\u95ee\u9898\u548c\u5c40\u9650\u6027\uff1a \u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u96be\u9898\uff0c\u5fae\u8f6f\u7814\u7a76\u9662\u79d1\u5b66\u667a\u80fd\u4e2d\u5fc3\u7684\u7814\u7a76\u5458\u4eec\u5c06\u7814\u7a76\u91cd\u70b9\u805a\u7126\u5728\u4e86\u5982\u4f55\u63d0\u9ad8\u5206\u5b50\u53ef\u89e3\u91ca\u6027\u3001\u964d\u4f4e\u8ba1\u7b97\u6210\u672c\u4ee5\u53ca\u8bc4\u4f30\u5b9e\u9645\u5e94\u7528\u51e0\u4e2a\u65b9\u9762\uff0c\u5e76\u521b\u65b0\u6027\u5730\u63d0\u51fa\u4e86\u901a\u7528\u5206\u5b50\u7ed3\u6784\u5efa\u6a21\u7f51\u7edc ViSNet (Vector-Scalar interactive graph neural Network)\u3002\u76f8\u5173\u6587\u7ae0\u201cEnhancing geometric representations for molecules with equivariant vector-scalar interactive message passing\u201d\u5df2\u53d1\u8868\u5728\u300a\u81ea\u7136-\u901a\u8baf\u300b\uff08Nature Communications\uff09\u6742\u5fd7\u4e0a\uff0c\u5e76\u540c\u65f6\u5165\u9009\u4e86\u201cAI and machine learning\u201d\u548c\u201cBiotechnology and method\u201d\u4e24\u4e2a\u9886\u57df\u7684\u7f16\u8f91\u7cbe\u9009\u6587\u7ae0\u3002 ViSNet\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/www.nature.com\/articles\/s41467-023-43720-2 (opens in new tab) \u201cAI and machine learning\u201d\u9886\u57df\u7f16\u8f91\u7cbe\u9009\u6587\u7ae0\u94fe\u63a5\uff1ahttps:\/\/www.nature.com\/collections\/ceiajcdbeb (opens in new tab) \u201cBiotechnology and method\u201d\u9886\u57df\u7f16\u8f91\u7cbe\u9009\u6587\u7ae0\u94fe\u63a5\uff1ahttps:\/\/www.nature.com\/collections\/idhhgedgig (opens in new tab) \u7814\u7a76\u5458\u4eec\u6700\u521d\u8ba1\u5212\u901a\u8fc7\u6709\u6548\u4e14\u5145\u5206\u5730\u5229\u7528\u5206\u5b50\u7ed3\u6784\u7684\u9886\u57df\u77e5\u8bc6\u6765\u8bbe\u8ba1\u6a21\u578b\u3002\u7531\u4e8e\u7ecf\u5178\u5206\u5b50\u52a8\u529b\u5b66\uff08molecular dynamics, MD\uff09\u901a\u8fc7\u660e\u786e\u63cf\u8ff0\u52bf\u80fd\u51fd\u6570\u4e2d\u7684\u952e\u957f\u3001\u952e\u89d2\u548c\u4e8c\u9762\u89d2\u6765\u6a21\u62df\u5206\u5b50\u8fd0\u52a8\uff0c\u6240\u4ee5\u53d7\u7ecf\u5178 MD [&hellip;]<\/p>\n","protected":false},"author":42735,"featured_media":1077369,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-content-parent":1012650,"footnotes":""},"research-area":[],"msr-locale":[268881],"msr-post-option":[],"class_list":["post-1077171","msr-blog-post","type-msr-blog-post","status-publish","has-post-thumbnail","hentry","msr-locale-zh_cn"],"msr_assoc_parent":{"id":1012650,"type":"lab"},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1077171","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-blog-post"}],"author":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/users\/42735"}],"version-history":[{"count":3,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1077171\/revisions"}],"predecessor-version":[{"id":1077372,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1077171\/revisions\/1077372"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1077369"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1077171"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1077171"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1077171"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1077171"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}