[{"data":1,"prerenderedAt":310},["ShallowReactive",2],{"content-query-2rTKMdv9I6":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":304,"_id":305,"_source":306,"_file":307,"_stem":308,"_extension":309},"/news/zh/2026-1-12","zh",false,"","智启科学新范式：浙江大学张强团队联手昇思MindSpore，发布知识图谱驱动的SciToolAgent","推出的面向多学科的通用AI智能体SciToolAgent，实现了在生物、化学、材料等领域自动化调用和管理数百种科学工具","2026-1-12","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/07/25/199b735845bf4106b44b2035dc97bd39.png","news",{"type":14,"children":15,"toc":301},"root",[16,24,30,35,40,51,56,65,76,85,93,98,103,111,119,124,129,136,144,152,157,164,169,174,181,189,200,208,213,220,228,233,240,248,253,260,268,273,280,288,296],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"智启科学新范式浙江大学张强团队联手昇思mindspore发布知识图谱驱动的scitoolagent",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":29},"在刚刚圆满落幕的昇思人工智能框架峰会2025上，一项名为SciToolAgent的科研成果引发广泛关注。这不仅是学术界对“Agentic Science”的一次深度探索，也是自主创新全栈软硬件平台（昇思MindSpore+昇腾硬件）与顶尖科研团队协同创新的又一次成功实践。当自动化、可解释、安全的“数字科学家”走进实验室，科学发现的齿轮正加速转动。",{"type":17,"tag":25,"props":31,"children":32},{},[33],{"type":23,"value":34},"浙江大学张强老师团队推出的面向多学科的通用AI智能体SciToolAgent，实现了在生物、化学、材料等领域自动化调用和管理数百种科学工具，使先进科研工具既能服务于专家也能助力非专业人士完成复杂领域任务。相关研究论文《SciToolAgent: A Knowledge Graph-Driven Scientific Agent for Multi-Tool Integration》已发表于2025年7月的《Nature Computational Science》期刊。",{"type":17,"tag":25,"props":36,"children":37},{},[38],{"type":23,"value":39},"论文地址：",{"type":17,"tag":25,"props":41,"children":42},{},[43],{"type":17,"tag":44,"props":45,"children":49},"a",{"href":46,"rel":47},"https://www.nature.com/articles/s43588-025-00849-y",[48],"nofollow",[50],{"type":23,"value":46},{"type":17,"tag":25,"props":52,"children":53},{},[54],{"type":23,"value":55},"代码地址：",{"type":17,"tag":25,"props":57,"children":58},{},[59],{"type":17,"tag":44,"props":60,"children":63},{"href":61,"rel":62},"https://github.com/HICAI-ZJU/SciToolAgent",[48],[64],{"type":23,"value":61},{"type":17,"tag":66,"props":67,"children":69},"div",{"style":68},"text-align: center;",[70],{"type":17,"tag":71,"props":72,"children":75},"img",{"src":73,"style":74,"alt":7},"/category/information/technology-blogs/banner/2026-1-12/1.jpg","display: block;margin: 0 auto;max-width:70%",[],{"type":17,"tag":25,"props":77,"children":78},{},[79],{"type":17,"tag":80,"props":81,"children":82},"strong",{},[83],{"type":23,"value":84},"# 01",{"type":17,"tag":25,"props":86,"children":87},{},[88],{"type":17,"tag":80,"props":89,"children":90},{},[91],{"type":23,"value":92},"科学发现范式的演进",{"type":17,"tag":25,"props":94,"children":95},{},[96],{"type":23,"value":97},"科学发现范式随着时代不断发展，经历了四个主要阶段：从炼金术士的实验探索，到牛顿理论模型的建立，再到第一性原理的计算机仿真，如今在人工智能与大数据驱动下，开启了AI驱动科研的新范式。基于大语言模型的科研智能体应运而生，正是这一演进的最新体现。",{"type":17,"tag":25,"props":99,"children":100},{},[101],{"type":23,"value":102},"随着科研任务复杂度提升，化学合成、蛋白质设计等过程对专业工具的依赖日益增强。然而，专业工具种类繁多、操作复杂、门槛高，现有方法难以有效整合多种工具，且往往忽视实验安全性问题。针对这些痛点，张强老师团队致力于构建自动化、可解释且安全的科学智能体，实现化学、材料等领域专业工具的自动化规划与调用，降低复杂科学工作的使用门槛。",{"type":17,"tag":25,"props":104,"children":105},{},[106],{"type":17,"tag":80,"props":107,"children":108},{},[109],{"type":23,"value":110},"# 02",{"type":17,"tag":25,"props":112,"children":113},{},[114],{"type":17,"tag":80,"props":115,"children":116},{},[117],{"type":23,"value":118},"SciToolAgent的核心创新",{"type":17,"tag":25,"props":120,"children":121},{},[122],{"type":23,"value":123},"研究团队提出的SciToolAgent包含Planner、Executor和Summarizer三个核心模块，通过结构化的科学工具知识图谱SciToolKG，整合了包括通用工具、BLAST、RDKit、ESMFold等在内的500+科学工具，覆盖化学、材料和生物等多个科学领域。并专门设有基于检索的毒性/爆炸性检测安全模块，保障实验过程的安全性。",{"type":17,"tag":25,"props":125,"children":126},{},[127],{"type":23,"value":128},"SciToolAgent通过基于SciToolKG的图检索增强生成机制，实现工具的智能化选择与执行。在团队构建的专用评测集上，其整体表现显著优于现有方法，并在蛋白质设计与分析、化学反应性预测、化学合成及MOF材料筛选等实际案例中，进一步验证了该系统在自动化复杂科学工作流方面的能力。",{"type":17,"tag":66,"props":130,"children":131},{"style":68},[132],{"type":17,"tag":71,"props":133,"children":135},{"src":134,"style":74,"alt":7},"/category/information/technology-blogs/banner/2026-1-12/2.jpg",[],{"type":17,"tag":25,"props":137,"children":138},{},[139],{"type":17,"tag":80,"props":140,"children":141},{},[142],{"type":23,"value":143},"# 03",{"type":17,"tag":25,"props":145,"children":146},{},[147],{"type":17,"tag":80,"props":148,"children":149},{},[150],{"type":23,"value":151},"性能表现卓越",{"type":17,"tag":25,"props":153,"children":154},{},[155],{"type":23,"value":156},"团队构建了自建评测集SciToolEval，包含531个科学问题（一级任务152项，二级任务379项）。从任务成功率、工具规划准确率和最终答案准确率三项指标评估，SciToolAgent表现显著优于现有方法。",{"type":17,"tag":66,"props":158,"children":159},{"style":68},[160],{"type":17,"tag":71,"props":161,"children":163},{"src":162,"style":74,"alt":7},"/category/information/technology-blogs/banner/2026-1-12/3.jpg",[],{"type":17,"tag":25,"props":165,"children":166},{},[167],{"type":23,"value":168},"依托昇思MindSpore AI框架的全面支持，SciToolAgent实现了高效开发与部署。该框架不仅提供了便捷的数据转换脚本，可自动完成对大模型输入数据的编码与填充，还原生支持热点模型的训练与推理，支持一键启动，大幅降低了使用门槛。通过与昇腾硬件的深度协同，系统充分发挥出自主创新算力的极致性能。",{"type":17,"tag":25,"props":170,"children":171},{},[172],{"type":23,"value":173},"在此强大基础上，SciToolAgent的核心模型仅需单机8卡配置，结合多维混合并行技术，便可在9小时内完成3万余条数据、多轮epoch的高效微调。基于SciToolKG数据微调后的Qwen2.5-7B模型，性能提升显著（+10%），已接近更大规模模型的表现。这不仅验证了知识图谱增强策略的有效性，更凸显出基于自主创新全栈软硬件进行模型定向优化的巨大潜力。",{"type":17,"tag":66,"props":175,"children":176},{"style":68},[177],{"type":17,"tag":71,"props":178,"children":180},{"src":179,"style":74,"alt":7},"/category/information/technology-blogs/banner/2026-1-12/4.jpg",[],{"type":17,"tag":25,"props":182,"children":183},{},[184],{"type":17,"tag":80,"props":185,"children":186},{},[187],{"type":23,"value":188},"# 04",{"type":17,"tag":25,"props":190,"children":191},{},[192],{"type":17,"tag":80,"props":193,"children":194},{},[195],{"type":17,"tag":80,"props":196,"children":197},{},[198],{"type":23,"value":199},"应用案例展示",{"type":17,"tag":25,"props":201,"children":202},{},[203],{"type":17,"tag":80,"props":204,"children":205},{},[206],{"type":23,"value":207},"1、蛋白质设计与分析",{"type":17,"tag":25,"props":209,"children":210},{},[211],{"type":23,"value":212},"蛋白质设计在药物发现、酶工程和合成生物学等生物学研究方中中具有重要意义。SciToolAgent通过自主调度和组合所需工具，能够完成包含序列设计，稳定性预测，结构预测，频率分析和二级结构验证等多个步骤的蛋白质设计任务。",{"type":17,"tag":66,"props":214,"children":215},{"style":68},[216],{"type":17,"tag":71,"props":217,"children":219},{"src":218,"style":74,"alt":7},"/category/information/technology-blogs/banner/2026-1-12/5.jpg",[],{"type":17,"tag":25,"props":221,"children":222},{},[223],{"type":17,"tag":80,"props":224,"children":225},{},[226],{"type":23,"value":227},"2、化学反应性预测",{"type":17,"tag":25,"props":229,"children":230},{},[231],{"type":23,"value":232},"化学反应性的预测是药物设计和有机合成中的关键环节。SciToolAgent基于不同分子特征(RDK, Morgan, charge descriptors) 对不同化学底物的反应性进行预测，并能够进一步识别该任务中最优的机器学习算法 (MLP, RF等) ，自动规划执行并选择分子特征和机器学习算法的最佳组合以提高预测准确率。",{"type":17,"tag":66,"props":234,"children":235},{"style":68},[236],{"type":17,"tag":71,"props":237,"children":239},{"src":238,"style":74,"alt":7},"/category/information/technology-blogs/banner/2026-1-12/6.jpg",[],{"type":17,"tag":25,"props":241,"children":242},{},[243],{"type":17,"tag":80,"props":244,"children":245},{},[246],{"type":23,"value":247},"3、化学合成与安全评估",{"type":17,"tag":25,"props":249,"children":250},{},[251],{"type":23,"value":252},"化学合成是化学研究的核心基础，在新型化合物设计与制备中发挥着关键作用。SciToolAgent可以自动完成合成路线预测 + 分子描述 + 专利查询 + 安全性分析全链条的化学合成与分析流程，其中内置的安全模块可检测有毒产物，确保实验安全，解决了科研实践中的关键隐患。",{"type":17,"tag":66,"props":254,"children":255},{"style":68},[256],{"type":17,"tag":71,"props":257,"children":259},{"src":258,"style":74,"alt":7},"/category/information/technology-blogs/banner/2026-1-12/7.jpg",[],{"type":17,"tag":25,"props":261,"children":262},{},[263],{"type":17,"tag":80,"props":264,"children":265},{},[266],{"type":23,"value":267},"4、MOF材料筛选",{"type":17,"tag":25,"props":269,"children":270},{},[271],{"type":23,"value":272},"MOF材料作为2025年诺贝尔化学奖获奖成果，在气体储存、催化、药物传输与分离等领域应用广泛。SciToolAgent通过自动筛选热稳定性、吸附量与价格等关键条件，大幅简化了MOF材料的设计流程，并依托多模型与数据库的智能调用，实现了高效的材料筛选与推荐。",{"type":17,"tag":66,"props":274,"children":275},{"style":68},[276],{"type":17,"tag":71,"props":277,"children":279},{"src":278,"style":74,"alt":7},"/category/information/technology-blogs/banner/2026-1-12/8.jpg",[],{"type":17,"tag":25,"props":281,"children":282},{},[283],{"type":17,"tag":80,"props":284,"children":285},{},[286],{"type":23,"value":287},"# 05",{"type":17,"tag":25,"props":289,"children":290},{},[291],{"type":17,"tag":80,"props":292,"children":293},{},[294],{"type":23,"value":295},"MindSpore+科研团队共创智能未来",{"type":17,"tag":25,"props":297,"children":298},{},[299],{"type":23,"value":300},"SciToolAgent的发布是昇思MindSpore生态建设的重要成果。通过顶尖高校与自主创新AI框架的“联创”，将前沿算法在硬核底座上实现优化。当顶尖的学术智慧碰撞自主创新的AI框架，我们看到的不仅是一个模型的诞生，更是自主创新AI助力全人类科学探索的无限可能。",{"title":7,"searchDepth":302,"depth":302,"links":303},4,[],"markdown","content:news:zh:2026-1-12.md","content","news/zh/2026-1-12.md","news/zh/2026-1-12","md",1776506060767]