[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fZ2NiKhUPK7cOru0ZTlgXfC5dVxvG3miODDDFzm-pRNo":3},{"item":4,"storyV2":86},{"slug":5,"type":6,"track":7,"season":8,"date":9,"featured":10,"teamName":11,"institution":13,"projectTitle":15,"summary":18,"body":20,"tags":24,"aiTags":27,"wetlabTags":30,"awards":33,"logoUrl":36,"materials":37,"gallery":38,"editorial":61,"cover":84},"njtech-syncar","image","T3","2026","2026-01-01",false,{"zh":12,"en":12},"NJTech-SynCAR",{"zh":14,"en":14},"南京工业大学生物与制药工程学院",{"zh":16,"en":17},"SynCAR —— 丁二胺合成关键酶改造","SynCAR — engineering a key enzyme for putrescine synthesis",{"zh":19,"en":19},"融合半理性设计与 AI 闭环进化，提升羧酸还原酶 MAB2962 对 GABA 的催化效率",{"zh":21,"en":23},[22],"南京工业大学 NJTech-SynCAR 盯着羧酸还原酶（CAR）：先用半理性设计筛出优秀变体，再把突变位置、方式和酶活数据交给蛋白质语言模型学习，让它预测下一批更有潜力的方案。",[22],{"zh":25,"en":26},[],[],{"zh":28,"en":29},[],[],{"zh":31,"en":32},[],[],{"zh":34,"en":35},[],[],"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fcd89b84259b0e8441120bfcb9",{},[39,47,54],{"imageUrl":40,"alt":41,"title":43,"description":45,"position":46},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fc6ba3e0a440f2bbc375bfdecb",{"zh":42,"en":42},"五位成员在南京生物工程技术研究中心前合影。",{"zh":44,"en":44},"NJTech-SynCAR 团队",{"zh":42,"en":42},"center center",{"imageUrl":48,"alt":49,"title":51,"description":53,"position":46},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fc0de4910797acc01e5219f7aa",{"zh":50,"en":50},"团队使用电脑和打印记录讨论复筛数据，确定下一轮候选与验证安排。",{"zh":52,"en":52},"复筛结果讨论",{"zh":50,"en":50},{"imageUrl":55,"alt":56,"title":58,"description":60,"position":46},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fc512e8d87b71798d79de31378",{"zh":57,"en":57},"成员围绕工作站查看计算结果，并与已有突变活性记录逐项核对。",{"zh":59,"en":59},"模型结果与实验数据对照",{"zh":57,"en":57},{"proposition":62,"listEvidence":63},{"zh":19,"en":19},[64,69,74,79],{"value":65,"label":67},{"zh":66,"en":66},"5 人",{"zh":68,"en":68},"成员人数",{"value":70,"label":72},{"zh":71,"en":71},"蔡一南",{"zh":73,"en":73},"队长",{"value":75,"label":77},{"zh":76,"en":76},"羧酸还原酶改造",{"zh":78,"en":78},"项目对象",{"value":80,"label":82},{"zh":81,"en":81},"酶活与交叉验证",{"zh":83,"en":83},"验证方式",{"tone":85,"heroImageUrl":40,"heroImagePosition":46},"gold",{"version":87,"publicSlug":5,"identity":88,"proposition":92,"lead":93,"listEvidence":94,"narrativeTendency":107,"hero":108,"presentation":113,"chapters":132},2,{"teamName":89,"projectTitle":90,"institution":91,"track":7},{"zh":12,"en":12},{"zh":16,"en":17},{"zh":14,"en":14},{"zh":19,"en":19},{"zh":22,"en":22},[95,98,101,104],{"value":96,"label":97},{"zh":66,"en":66},{"zh":68,"en":68},{"value":99,"label":100},{"zh":71,"en":71},{"zh":73,"en":73},{"value":102,"label":103},{"zh":76,"en":76},{"zh":78,"en":78},{"value":105,"label":106},{"zh":81,"en":81},{"zh":83,"en":83},"evidence-led",{"role":109,"title":110,"caption":111,"alt":112,"position":46,"imageUrl":40},"TEAM",{"zh":44,"en":44},{"zh":42,"en":42},{"zh":42,"en":42},{"template":114,"variant":115,"logo":116,"openingMedia":123},"documentary-profile","rich",{"role":117,"title":118,"caption":120,"alt":122,"position":46,"imageUrl":36},"LOGO",{"zh":119,"en":119},"NJTech-SynCAR队徽",{"zh":121,"en":121},"队徽：CAR 三个字母被拼成一辆车，车身是 DNA 双螺旋，上方一串二进制。CAR 既是羧酸还原酶（carboxylic acid reductase），也是那辆车。",{"zh":121,"en":121},[124,128],{"role":109,"title":125,"caption":126,"alt":127,"position":46,"imageUrl":48},{"zh":52,"en":52},{"zh":50,"en":50},{"zh":50,"en":50},{"role":109,"title":129,"caption":130,"alt":131,"position":46,"imageUrl":55},{"zh":59,"en":59},{"zh":57,"en":57},{"zh":57,"en":57},[133,142,176,202,242,265],{"id":134,"eyebrow":135,"title":137,"body":139,"theme":141,"type":141},"project-background",{"zh":136,"en":136},"项目背景",{"zh":138,"en":138},"先说清楚 CAR 是干什么的",{"zh":140,"en":140},"酶就像大自然提供的「高效催化剂」，能让工业生产更快、更环保。它的本事是把羧酸转化为醛 —— 而醛类化合物是医药、香料和新能源领域的常用原料。这条路线在技术路线图上画得很清楚：L-谷氨酸 → γ-氨基丁酸（GABA）→ 1,4-丁二胺。最后那个 1,4-丁二胺，是尼龙-46 等聚合物的单体，目前主要靠石化路线生产。把它换成生物制造，卡点就落在中间这一步的酶效率上。队徽把这件事变成了一个双关：CAR 三个字母被拼成一辆车，车身是 DNA 双螺旋。","problem",{"id":143,"eyebrow":144,"title":146,"body":148,"theme":150,"type":151,"facts":152,"image":168},"research-context",{"zh":145,"en":145},"研究路径",{"zh":147,"en":147},"两条腿：先半理性，再交给模型",{"zh":149,"en":149},"路线图的第一个模块是半理性设计 —— 序列分析、结构预测、构建突变库，用传统方法先拿到一批可靠的数据点。第二个模块才是 AI。把突变位置、方式和酶活性等数据交给人工智能学习，AI 总结规律、预测更有潜力的改造方案，再通过「实验—学习—再设计」的循环不断进化。顺序在这里是有意义的。蛋白质语言模型需要有标注的活性数据才能微调，而这批数据只能来自实验台。所以不是「AI 先给答案、实验去验证」，而是先有一轮实打实的半理性筛选，模型才有东西可学。复筛结果图就是这批数据的样子：几十个突变体按 PSSM 归一化后的表现排序，从最高一路排到最低。","source","source-context",[153,158,163],{"label":154,"value":156},{"zh":155,"en":155},"目标酶",{"zh":157,"en":157},"MAB2962",{"label":159,"value":161},{"zh":160,"en":160},"改造位点",{"zh":162,"en":162},"342、514、281、420",{"label":164,"value":166},{"zh":165,"en":165},"模型底座",{"zh":167,"en":167},"ESM2（35M、150M、650M 比选）",{"role":169,"title":170,"caption":172,"alt":174,"position":46,"imageUrl":175},"RESULT",{"zh":171,"en":171},"复筛结果",{"zh":173,"en":173},"以 PSSM 归一化后的各突变体表现，按高低排序。横轴每一格是一个突变。",{"zh":173,"en":173},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fcdd0210256e95c27a283f07a9",{"id":177,"eyebrow":178,"title":180,"theme":182,"type":183,"semantics":184,"steps":185},"project-process",{"zh":179,"en":179},"项目过程",{"zh":181,"en":181},"项目推进与验证路径","process","process-flow","hybrid",[186,195],{"order":187,"phase":188,"title":190,"body":192,"status":194},1,{"zh":189,"en":189},"阶段 1",{"zh":191,"en":191},"位点是这么一个一个试出来的",{"zh":193,"en":193},"柱状图记录了 342 位单点突变，以及以 D281P 为亲本继续组合 514 位突变的实验比较。部分突变体的细胞催化活力高于相应亲本。","described",{"order":87,"phase":196,"title":198,"body":200,"status":194},{"zh":197,"en":197},"阶段 2",{"zh":199,"en":199},"模型选型是比出来的，不是挑出来的",{"zh":201,"en":201},"底座是 esm2_t33_650M_UR50D，33 层，隐藏维度 1280；冻结 0–21 层（共 22 层，占 66.7%），总参数 651.0M，其中可训练 218.1M。训练集 105 条、测试集 35 条，做四折交叉验证。四折的 Spearman 相关系数分别是 0.618、0.705、0.455、0.227。两组数放在一起看，信息量比任何一个单独的数字都大：折与折之间的方差很明显，说明在这个数据量级上，结果对数据划分相当敏感。",{"id":203,"eyebrow":204,"title":206,"theme":208,"type":209,"items":210},"project-figure",{"zh":205,"en":205},"项目证据",{"zh":207,"en":207},"项目图表与方法记录","evidence","figures",[211,219,227,235],{"role":212,"title":213,"caption":215,"alt":217,"position":46,"imageUrl":218},"DRY_LAB",{"zh":214,"en":214},"训练日志",{"zh":216,"en":216},"底座模型 esm2_t33_650M_UR50D，33 层、隐藏维度 1280，冻结 0–21 层（占 66.7%），总参数 651.0M，可训练 218.1M。四折交叉验证的 Spearman 分别是 0.618、0.705、0.455、0.227。",{"zh":216,"en":216},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fc41e5c366be99c6d3adf5dd4a",{"role":220,"title":221,"caption":223,"alt":225,"position":46,"imageUrl":226},"WET_LAB",{"zh":222,"en":222},"通风橱内样品处理",{"zh":224,"en":224},"队员完成移液和反应体系配置，为羧酸还原酶突变体的活性测定准备样品。",{"zh":224,"en":224},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fcd560bf395b7995b1f9b29652",{"role":228,"title":229,"caption":231,"alt":233,"position":46,"imageUrl":234},"METHOD",{"zh":230,"en":230},"SynCAR 技术路线",{"zh":232,"en":232},"路线图串联半理性突变、AI 模型迭代、平台化 CAR 设计与后续实验评价。",{"zh":232,"en":232},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fc060bd5a6d1a2cf3efbd82a33",{"role":169,"title":236,"caption":238,"alt":240,"position":46,"imageUrl":241},{"zh":237,"en":237},"342 位的饱和突变",{"zh":239,"en":239},"横轴从野生型（WT）依次排开 A342C、A342N、A342H、A342Y、A342K…，纵轴是细胞催化活力（A₄₃₅ₙₘ）。同一个位点换成不同氨基酸，活力能差三四倍。",{"zh":239,"en":239},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fc75b14b97707ea08088636b46",{"id":243,"eyebrow":244,"title":246,"theme":248,"type":249,"items":250},"team-gallery",{"zh":245,"en":245},"团队协作",{"zh":247,"en":247},"计算、实验与候选复核记录","people","gallery",[251,258],{"role":212,"title":252,"caption":254,"alt":256,"position":46,"imageUrl":257},{"zh":253,"en":253},"突变记录结构化处理",{"zh":255,"en":255},"脚本读取表格中的突变记录，分别整理单点与双点突变，并将活性值与候选序列逐条对应。",{"zh":255,"en":255},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fc1314b59bb9bcc978a2f941cc",{"role":220,"title":259,"caption":261,"alt":263,"position":46,"imageUrl":264},{"zh":260,"en":260},"酶标仪读取活性",{"zh":262,"en":262},"队员在酶标仪前读取多孔板结果，将候选突变体的实验数据汇入后续分析。",{"zh":262,"en":262},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fce16e97bb16b3d3f10673ea2f",{"id":266,"eyebrow":267,"title":269,"theme":248,"type":271,"items":272},"team-stories",{"zh":268,"en":268},"故事节点",{"zh":270,"en":270},"项目过程与团队记录","stories",[273,287],{"label":274,"title":276,"body":278,"image":280},{"zh":275,"en":275},"节点 1",{"zh":277,"en":277},"屏幕上的另一半：结构与接触",{"zh":279,"en":279},"一张里，左边是蛋白的三维结构（蓝色卡通模型），右边是残基间的接触热图，两者并排；另一张是同一套界面的深色模式，下方带着一张数据表。还有一张是代码：把 Excel 里的突变记录解析成结构化数据 —— 用正则把「A123B」这样的写法拆成位点与替换氨基酸，单点与双点突变分开处理，活性值逐条对上号。这类脚本不出现在任何成果图里，但它决定了模型到底吃进去了什么。",{"role":212,"title":281,"caption":283,"alt":285,"position":46,"imageUrl":286},{"zh":282,"en":282},"结构与接触图并排看",{"zh":284,"en":284},"左边是蛋白三维结构，右边是残基间接触热图。",{"zh":284,"en":284},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fc0067212ed339083f07649965",{"label":288,"title":290,"body":292,"image":294},{"zh":289,"en":289},"节点 2",{"zh":291,"en":291},"从单个突变体到平台化 CAR 设计",{"zh":293,"en":293},"路线图的第三个模块叫平台化 CAR 设计：CAR 通用突变、知识迁移、SynCAR 平台。改一个酶，做完就结束了；让模型学会改一类酶，意味着这次积累的突变—活性数据、这套微调流程、这份可迁移的设计流程，本身就是产出。第四个模块列出了预期目标：高活性羧酸还原酶变体、AI 驱动的 CAR 设计能力、丁二胺绿色制造、可迁移的设计流程、更低成本的 CAR 反应。到那时，这套平台需要拿出的不只是一个更好的数字，而是它确实迁移过一次的证据。",{"role":220,"title":295,"caption":297,"alt":299,"position":46,"imageUrl":300},{"zh":296,"en":296},"九十六孔板加样",{"zh":298,"en":298},"多孔板用于并行比较不同突变体的细胞催化活力，形成模型训练与复筛所需的数据。",{"zh":298,"en":298},"\u002Fapi\u002Fteam-showcase\u002Fnjtech-syncar\u002Fmedia\u002Fc3da6ef8bd7c8ad1cc554c924"]