[{"data":1,"prerenderedAt":175},["ShallowReactive",2],{"content-query-ofKpqIchGW":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":169,"_id":170,"_source":171,"_file":172,"_stem":173,"_extension":174},"/news/zh/3762","zh",false,"","Nature子刊重磅发布！昇思MindSpore使能全球首个亿级细胞大模型诞生","中山大学、重庆大学等科研院所近日在《Nature Communications》发表开创性研究，正式推出基于昇思MindSpore AI框架与昇腾AI硬件训练的CellFM模型——全球首个预训练规模达1亿人类细胞的转录组学基础模型。这标志着AI技术在生命科学领域取得革命性突破。","2025-06-16","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/06/20/a59fa497b94b4fe7a32a6ab48f73fd97.png","news",{"type":14,"children":15,"toc":166},"root",[16,24,29,37,46,54,59,66,75,80,87,95,100,108,116,121,126,131,136,141,152],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"nature子刊重磅发布昇思mindspore使能全球首个亿级细胞大模型诞生",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":9},{"type":17,"tag":25,"props":30,"children":31},{},[32],{"type":17,"tag":33,"props":34,"children":36},"img",{"alt":7,"src":35},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/06/20/e3257ab621c341d7b2067407d292488e.png",[],{"type":17,"tag":25,"props":38,"children":39},{},[40],{"type":17,"tag":41,"props":42,"children":43},"strong",{},[44],{"type":23,"value":45},"# 01",{"type":17,"tag":25,"props":47,"children":48},{},[49],{"type":17,"tag":41,"props":50,"children":51},{},[52],{"type":23,"value":53},"攻克****单细胞研究三大难关",{"type":17,"tag":25,"props":55,"children":56},{},[57],{"type":23,"value":58},"面对单细胞数据分析的噪声干扰、批次效应与数据稀疏性挑战，研究团队依托昇思MindSpore构建CellFM模型，实现对海量转录组数据的统一建模，为解码细胞异质性提供全新范式。",{"type":17,"tag":25,"props":60,"children":61},{},[62],{"type":17,"tag":33,"props":63,"children":65},{"alt":7,"src":64},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/06/20/6be9946d8dcf4ac99f0a7dfd6b6603eb.png",[],{"type":17,"tag":25,"props":67,"children":68},{},[69],{"type":17,"tag":70,"props":71,"children":72},"em",{},[73],{"type":23,"value":74},"图1： 预训练的CellFM模型适用于多种单细胞下游分析，包括细胞类型注释、扰动预测、基因网络推断和基因功能预测。",{"type":17,"tag":25,"props":76,"children":77},{},[78],{"type":23,"value":79},"在技术创新层面，CellFM通过三大核心突破构建了坚实的技术壁垒。首先，该模型实现了超大规模预训练，成功建成全球首个基于一亿人类细胞训练的转录组基础模型，参数量级突破八亿大关，创造了该领域新的技术标杆。其次，在模型架构方面进行关键革新：采用改进型RetNet框架作为基础结构，通过优化注意力机制显著提升特征提取能力，并开发出高效的数据处理流程以应对海量生物数据。再者，该模型展现了跨领域的卓越性能表现，包括高精度细胞类型注释能力、精准的扰动预测性能，以及对基因功能及其相互作用关系的深度解析能力，为多学科研究提供有力支持。",{"type":17,"tag":25,"props":81,"children":82},{},[83],{"type":17,"tag":33,"props":84,"children":86},{"alt":7,"src":85},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2025/06/20/59d82848df614492ab72eacee0679725.png",[],{"type":17,"tag":25,"props":88,"children":89},{},[90],{"type":17,"tag":70,"props":91,"children":92},{},[93],{"type":23,"value":94},"图2：a. CellFM框架概述。CellFM由嵌入模块、ERetNet模块和LoRA模块组成。b. 每层ERetNet模块集成了MHA、SGLU和LN单元。",{"type":17,"tag":25,"props":96,"children":97},{},[98],{"type":23,"value":99},"作为生物医学领域的新型智能基础研究，CellFM正在构建多维应用价值体系。在基础研究方面，有力支撑细胞鉴定、细胞状态分析与基因调控网络研究等核心科研方向；在医学实践领域，服务于疾病发生机制解析、药物响应预测及个性化诊疗方案制定；在生物工程应用维度，有效赋能细胞改造工程、基因治疗技术创新与新药研发进程。",{"type":17,"tag":25,"props":101,"children":102},{},[103],{"type":17,"tag":41,"props":104,"children":105},{},[106],{"type":23,"value":107},"# 02",{"type":17,"tag":25,"props":109,"children":110},{},[111],{"type":17,"tag":41,"props":112,"children":113},{},[114],{"type":23,"value":115},"昇思MindSpore AI框架三大核心技术赋能",{"type":17,"tag":25,"props":117,"children":118},{},[119],{"type":23,"value":120},"在高效分布式训练方面，昇思MindSpore通过自动并行策略实现对8亿参数规模的精准适配，结合数据并行、模型并行与流水线并行的多维协同优化机制，在保证开发灵活性的同时借助动静态图融合技术达成极致性能表现。",{"type":17,"tag":25,"props":122,"children":123},{},[124],{"type":23,"value":125},"在智能内存管理领域，昇思MindSpore图算融合技术有效优化了内存资源占用，梯度聚合优化方案显著降低了分布式训练中的通信开销，配合可动态适配超大规模模型的自动化内存调度系统，共同构建起高效内存管理生态。",{"type":17,"tag":25,"props":127,"children":128},{},[129],{"type":23,"value":130},"针对全链路性能优化，昇思MindSpore AI框架集成混合精度训练功能实现FP16与FP32精度的智能切换，依托算子融合引擎减少冗余内存访问操作，并通过多层次的图层编译优化策略全面提升计算效率。这三重技术体系形成紧密协同，为超大规模AI模型提供了坚实的底层支撑。",{"type":17,"tag":25,"props":132,"children":133},{},[134],{"type":23,"value":135},"CellFM的成功实践充分验证了昇思MindSpore在使能生物大模型领域的技术领导力。未来，昇思MindSpore将持续助力科研团队推动模型进化，拓展多物种细胞支持范围、实现多组学数据融合分析、丰富下游任务应用场景以及持续优化模型计算效能。",{"type":17,"tag":25,"props":137,"children":138},{},[139],{"type":23,"value":140},"访问昇思MindSpore官网，了解昇思MindSpore AI4S相关技术：",{"type":17,"tag":25,"props":142,"children":143},{},[144],{"type":17,"tag":145,"props":146,"children":150},"a",{"href":147,"rel":148},"https://www.mindspore.cn",[149],"nofollow",[151],{"type":23,"value":147},{"type":17,"tag":25,"props":153,"children":154},{},[155,157,164],{"type":23,"value":156},"[参考文献：[Zeng, Y., et al. CellFM: a large-scale foundation model pre-trained on transcriptomics of 100 million human cells. Nature Communications 16, 4679 (2025)](",{"type":17,"tag":145,"props":158,"children":161},{"href":159,"rel":160},"https://doi.org/10.1038/s41467-025-59926-5)%5C",[149],[162],{"type":23,"value":163},"https://doi.org/10.1038/s41467-025-59926-5)\\",{"type":23,"value":165},"]",{"title":7,"searchDepth":167,"depth":167,"links":168},4,[],"markdown","content:news:zh:3762.md","content","news/zh/3762.md","news/zh/3762","md",1776506088957]