[{"data":1,"prerenderedAt":133},["ShallowReactive",2],{"content-query-71nKft5EWm":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":127,"_id":128,"_source":129,"_file":130,"_stem":131,"_extension":132},"/news/zh/2023-3-28-3","zh",false,"","天工大模型","“天工”抗体设计模型是基于昇思MindSpore打造的一款创新工具，能够实现抗体功能设计、序列嫁接和活性预测等多种任务。","2023-3-28","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/25/199b735845bf4106b44b2035dc97bd39.png","news",{"type":14,"children":15,"toc":114},"root",[16,24,29,35,40,45,50,63,68,73,80,85,91,96,102,107],{"type":17,"tag":18,"props":19,"children":21},"element","h2",{"id":20},"项目说明",[22],{"type":23,"value":20},"text",{"type":17,"tag":18,"props":25,"children":27},{"id":26},"案例背景",[28],{"type":23,"value":26},{"type":17,"tag":30,"props":31,"children":32},"p",{},[33],{"type":23,"value":34},"抗体类药物已被广泛批准用于治疗多种疾病，具有巨大的社会价值，尤其在病毒引起的传染病等方面发挥着重要作用。抗体药物市场空间快速发展，然而，传统的抗体设计方法效率低下、成本高昂，难以满足日益增长的临床需求。",{"type":17,"tag":18,"props":36,"children":38},{"id":37},"案例简介",[39],{"type":23,"value":37},{"type":17,"tag":30,"props":41,"children":42},{},[43],{"type":23,"value":44},"针对这些需求与挑战，昌平实验室联合昇思 MindSpore 打造的“天工”抗体设计模型应运而生。它是一种能够与实验进行交互、具备生物先验知识的生成式 AI，能够实现对非人源有效抗体的人源化，从而降低抗体实验研发成本。此外，它还能进一步优化具有功效的抗体的可成药性、降低生产成本，最终实现对可成药抗体的高效增值。它将为抗体药物的研发带来革命性的变革。",{"type":17,"tag":30,"props":46,"children":47},{},[48],{"type":23,"value":49},"“天工”抗体设计模型是基于昇思 MindSpore 打造的一款创新工具，能够实现抗体功能设计、序列嫁接和活性预测等多种任务。天工模型设计效率相比传统抗体设计方法提升一个数量级以上，基于天工模型嫁接改造得到的人源抗体，经协和医学院团队验证抗体表达量提升约 5 至 10 倍，有望将抗体生产成本降低 2 倍以上。",{"type":17,"tag":51,"props":52,"children":55},"div",{"className":53},[54],"tiangong-img-box",[56,58],{"type":23,"value":57},"\n    ",{"type":17,"tag":59,"props":60,"children":62},"img",{"src":61},"/category/information/news/banner/2023-3-28-3.png",[],{"type":17,"tag":30,"props":64,"children":65},{},[66],{"type":23,"value":67},"“天工”抗体设计模型是业界多领域交叉融合的杰出成果。依托昇腾硬件与昇思 MindSpore AI 框架的基础软硬件能力， “天工”模型实现高效训练和推理。昌平实验室则为该模型注入了抗体设计领域的生物先验知识，使其能够同时支持高精度的功能设计和嫁接设计。协和医学院以其丰富的临床经验，迅速对新设计的抗体进行实验验证，并将湿实验结果反馈至模型，形成了干湿实验的闭环，为“天工”模型的完善和应用提供了强有力的支持。",{"type":17,"tag":18,"props":69,"children":71},{"id":70},"案例优势",[72],{"type":23,"value":70},{"type":17,"tag":74,"props":75,"children":77},"h3",{"id":76},"_1软件协同高效内存复用模型性能提升",[78],{"type":23,"value":79},"1、软件协同：高效内存复用，模型性能提升",{"type":17,"tag":30,"props":81,"children":82},{},[83],{"type":23,"value":84},"昇思 MindSpore 结合框架高效的内存复用能力，使用了重计算技术对模型进行优化。该技术网络正向执行是释放激活层的计算结果，在模型反向计算时重新计算正向激活层的结果，有效降低峰值内存，最终将训练内存减少了 10 倍。lazy inline 编译优化技术，能标记循环体中频繁调用的子图，推迟 inline 处理至编译的最后阶段，大幅度缩短了编译时间，将“天工”模型的编译性能提升了 5 倍以上。",{"type":17,"tag":74,"props":86,"children":88},{"id":87},"_2高效设计降低研发成本加速临床应用",[89],{"type":23,"value":90},"2、高效设计：降低研发成本，加速临床应用",{"type":17,"tag":30,"props":92,"children":93},{},[94],{"type":23,"value":95},"传统抗体研发投入成本高，导致许多抗体开发商长期亏损，而人源抗体的湿实验成本也相对昂贵。“天工”模型通过多段功能区域的联合分布改造生成抗体序列，能够高效筛选出具有优良活性的抗体候选。经过实验验证，经过“天工”模型嫁接改造的抗体相较于原始抗体，表达产率提升了约 5~10 倍。这一显著提升的表达产率为生产环节带来了巨大的效益。表达产率的提升意味着在相同的生产周期内，可以获得更多的抗体产品，从而降低了单位抗体的生产成本。",{"type":17,"tag":74,"props":97,"children":99},{"id":98},"_3智能预测提升抗体活性优化治疗效果",[100],{"type":23,"value":101},"3、智能预测：提升抗体活性，优化治疗效果",{"type":17,"tag":30,"props":103,"children":104},{},[105],{"type":23,"value":106},"“天工”模型还具备强大的活性预测能力，能够准确预测抗体的生物活性，为优化抗体结构、提升治疗效果提供了有力依据。通过智能预测，研究人员可以更加精准地筛选出具有优良活性的抗体，提高药物研发的成功率。针对一个具有潜在癌症治疗功效的鼠源抗体选择作为人源化目标，通过“天工”模型进行嫁接后得到的抗体序列，相比于原始序列在人源性得分上达到了 70%以上的提升，人源性分数表示人源化后的抗体与人类抗体序列的一致性程度，通常较高的人源性分数表示临床免疫原性更低、稳定性更高。",{"type":17,"tag":108,"props":109,"children":111},"style",{"lang":110},"scss",[112],{"type":23,"value":113},"\n    .tiangong-img-box{\n        width: 100%;\n        padding: 24px;\n        background: #f5f9fb;\n        margin: 24px 0;\n        display: flex;\n        justify-content: center;\n\n        img {\n           width: 63%;\n        }\n    }\n",{"title":7,"searchDepth":115,"depth":115,"links":116},4,[117,119,120,121],{"id":20,"depth":118,"text":20},2,{"id":26,"depth":118,"text":26},{"id":37,"depth":118,"text":37},{"id":70,"depth":118,"text":70,"children":122},[123,125,126],{"id":76,"depth":124,"text":79},3,{"id":87,"depth":124,"text":90},{"id":98,"depth":124,"text":101},"markdown","content:news:zh:2023-3-28-3.md","content","news/zh/2023-3-28-3.md","news/zh/2023-3-28-3","md",1776506058756]