[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fhnjWFCU6VVJVI40glAyEHsv2Da2eIHOYGtvag2ZcAi8":3},{"item":4,"storyV2":76},{"slug":5,"type":6,"track":7,"season":8,"date":9,"featured":10,"teamName":11,"institution":13,"projectTitle":15,"summary":17,"body":19,"tags":23,"aiTags":26,"wetlabTags":29,"awards":32,"logoUrl":35,"materials":36,"gallery":37,"editorial":51,"cover":74},"tju-pichia","image","T5","2026","2026-01-01",false,{"zh":12,"en":12},"TJU-Pichia",{"zh":14,"en":14},"天津大学合成生物与生物制造学院",{"zh":16,"en":16},"虚拟细胞赋能蛋白高效表达",{"zh":18,"en":18},"比较模型生成与规则优化序列，并据第一轮结果设计融合优化管线",{"zh":20,"en":22},[21],"TJU-Pichia 比较模型生成和规则优化两条密码子优化路线，用同一套指标评估差异，并据第一轮结果确定下一轮约束。",[21],{"zh":24,"en":25},[],[],{"zh":27,"en":28},[],[],{"zh":30,"en":31},[],[],{"zh":33,"en":34},[],[],"\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fc7ef266032befc7e3ff592916",{},[38,45],{"imageUrl":39,"alt":40,"title":42,"description":43,"position":44},"\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fc48461510c9b6565cd48cdeb1",{"zh":41,"en":41},"TJU-Pichia队伍合影",{"zh":41,"en":41},{"zh":41,"en":41},"center center",{"imageUrl":46,"alt":47,"title":49,"description":50,"position":44},"\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fc849703e3195e9b340da9bb03",{"zh":48,"en":48},"TJU-Pichia队员讨论项目方案",{"zh":48,"en":48},{"zh":48,"en":48},{"proposition":52,"listEvidence":53},{"zh":18,"en":18},[54,59,64,69],{"value":55,"label":57},{"zh":56,"en":56},"5 人",{"zh":58,"en":58},"成员人数",{"value":60,"label":62},{"zh":61,"en":61},"贾子晗",{"zh":63,"en":63},"队长",{"value":65,"label":67},{"zh":66,"en":66},"毕赤密码子优化",{"zh":68,"en":68},"项目对象",{"value":70,"label":72},{"zh":71,"en":71},"生成指标对照",{"zh":73,"en":73},"验证方式",{"tone":75,"heroImageUrl":39,"heroImagePosition":44},"indigo",{"version":77,"publicSlug":5,"identity":78,"proposition":82,"lead":83,"listEvidence":84,"narrativeTendency":97,"hero":98,"presentation":101,"chapters":114},2,{"teamName":79,"projectTitle":80,"institution":81,"track":7},{"zh":12,"en":12},{"zh":16,"en":16},{"zh":14,"en":14},{"zh":18,"en":18},{"zh":21,"en":21},[85,88,91,94],{"value":86,"label":87},{"zh":56,"en":56},{"zh":58,"en":58},{"value":89,"label":90},{"zh":61,"en":61},{"zh":63,"en":63},{"value":92,"label":93},{"zh":66,"en":66},{"zh":68,"en":68},{"value":95,"label":96},{"zh":71,"en":71},{"zh":73,"en":73},"evidence-led",{"role":99,"alt":100,"position":44,"imageUrl":39},"TEAM",{"zh":41,"en":41},{"template":102,"variant":103,"logo":104,"openingMedia":111},"documentary-profile","rich",{"role":105,"title":106,"caption":108,"alt":110,"position":44,"imageUrl":35},"LOGO",{"zh":107,"en":107},"TJU-Pichia队徽",{"zh":109,"en":109},"队徽：一条蓝色缎带绕着一个白球 —— 螺旋缠住的，是它正要折出来的那个蛋白。",{"zh":109,"en":109},[112],{"role":99,"alt":113,"position":44,"imageUrl":46},{"zh":48,"en":48},[115,124,150,184,210,239],{"id":116,"eyebrow":117,"title":119,"body":121,"theme":123,"type":123},"project-background",{"zh":118,"en":118},"项目背景",{"zh":120,"en":120},"毕赤酵母密码子适配问题",{"zh":122,"en":122},"乳铁蛋白在婴幼儿食品和保健品领域需求旺盛，但天然提取产量有限。毕赤酵母是常用的乳铁蛋白表达宿主，但其密码子偏好与乳铁蛋白天然编码序列存在差异。同一个氨基酸可以由多个密码子编码，而每种宿主对这些同义密码子的偏好不同。毕赤酵母的密码子使用偏好与乳铁蛋白差异显著，导致产量低、稳定性差。密码子优化在不改变蛋白氨基酸序列的前提下，使编码序列适配宿主表达偏好。传统做法是规则替换：查出宿主最偏好的那个密码子，逐位换上去。项目引入序列生成模型，以同时考虑密码子偏好、序列长度与表达约束。","problem",{"id":125,"eyebrow":126,"title":128,"body":130,"theme":132,"type":133,"facts":134},"research-context",{"zh":127,"en":127},"研究路径",{"zh":129,"en":129},"模型生成与规则优化对照",{"zh":131,"en":131},"第一轮 DBTL 设置模型生成与规则优化两条对照路线。opt1_model_based：把同一条乳铁蛋白氨基酸序列输入 Pichia-CLM 预训练模型，推理生成优化后的 CDS 序列，不加任何后置规则过滤。opt2_rule_based：采用毕赤酵母经典的高频密码子替换规则，逐位替换。两条路线使用同一套指标与计算脚本进行对照。两套序列使用完全一致的计算代码，批量计算 GC 含量、CAI、ENC、稀有密码子统计与综合评分，消除工具误差；全部计算代码与原始序列 FASTA 文件归档 GitLab。项目团队还预先写下了预期结果 —— 规则优化会拿到更高的 CAI 和综合评分，模型序列更短但密码子多样性极低。","source","source-context",[135,140,145],{"label":136,"value":138},{"zh":137,"en":137},"目标宿主",{"zh":139,"en":139},"毕赤酵母",{"label":141,"value":143},{"zh":142,"en":142},"对照路线",{"zh":144,"en":144},"模型生成、规则优化",{"label":146,"value":148},{"zh":147,"en":147},"复核指标",{"zh":149,"en":149},"GC、CAI、ENC",{"id":151,"eyebrow":152,"title":154,"theme":156,"type":157,"semantics":158,"steps":159},"project-process",{"zh":153,"en":153},"项目过程",{"zh":155,"en":155},"项目推进与验证路径","process","process-flow","hybrid",[160,169,176],{"order":161,"phase":162,"title":164,"body":166,"status":168},1,{"zh":163,"en":163},"阶段 1",{"zh":165,"en":165},"两种优化路线的指标对比",{"zh":167,"en":167},"长度上，opt1 是 498 bp，opt2 是 2073 bp。同一条氨基酸序列的两种优化结果，长度差异源于密码子选择策略不同。GC 含量：opt1 全局 33.53%，opt2 为 37.39%。决定性的一组：CAI 0.8011 对 0.905；宿主高频偏好密码子占比 0.6% 对 30.68%；密码子阶段综合评分 31.92 对 76.04。纯 AI 模型自主生成的序列虽然能规避稀有密码子并达到基础 CAI 阈值，但未能充分贴合毕赤酵母内源密码子使用偏好；传统规则优化牺牲了序列长度，却大幅提升了各项核心指标。两组唯一打平的地方是稀有密码子 —— 都是 0，都能规避翻译提前终止的问题。","described",{"order":77,"phase":170,"title":172,"body":174,"status":168},{"zh":171,"en":171},"阶段 2",{"zh":173,"en":173},"模型生成与约束筛选融合",{"zh":175,"en":175},"Learn 阶段比较两条路线：模型生成序列较短、GC 较低，规则优化的密码子适配度更高；下一轮将融合模型生成与多约束筛选。下一轮将搭建模型生成与多约束后置筛选的融合管线。第一轮的失败，直接定义了第二轮的结构。",{"order":177,"phase":178,"title":180,"body":182,"status":168},3,{"zh":179,"en":179},"阶段 3",{"zh":181,"en":181},"跨学科协作与模型调试",{"zh":183,"en":183},"做算法的同学一开始和做实验的同学思路很难对齐，后期实验数据也一直达不到预期。团队采用坐在一起讨论、梳理问题、分工试错，一点点磨合。",{"id":185,"eyebrow":186,"title":188,"theme":190,"type":191,"items":192},"project-figure",{"zh":187,"en":187},"项目证据",{"zh":189,"en":189},"项目图表与方法记录","evidence","figures",[193,202],{"role":194,"title":195,"caption":197,"alt":199,"position":44,"imageUrl":201},"DRY_LAB",{"zh":196,"en":196},"模型与规则两条路线的起点",{"zh":198,"en":198},"同一条乳铁蛋白氨基酸序列被分别交给 Pichia-CLM 和高频密码子替换规则。队员在电脑前整理两组输出，随后用同一套代码评测，避免工具差异干扰这场对照。",{"zh":200,"en":200},"队员在电脑前整理密码子优化序列","\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fc592693a3a9415b4f4aceedf1",{"role":203,"title":204,"caption":206,"alt":208,"position":44,"imageUrl":209},"RESULT",{"zh":205,"en":205},"TJU-Pichia项目图表",{"zh":207,"en":207},"项目海报。",{"zh":207,"en":207},"\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fcca24bf01f29f30bbebc95959",{"id":211,"eyebrow":212,"title":214,"theme":216,"type":217,"items":218},"team-gallery",{"zh":213,"en":213},"团队协作",{"zh":215,"en":215},"计算、实验与候选复核记录","people","gallery",[219,227,235],{"role":203,"title":220,"caption":222,"alt":224,"position":44,"imageUrl":226},{"zh":221,"en":221},"从蛋白序列生成宿主适配 CDS",{"zh":223,"en":223},"架构图把训练与推理分成两步：模型同时学习氨基酸与编码序列的对应关系；面对目标蛋白时，再生成面向宿主表达的候选 CDS，交给后处理规则继续筛选。",{"zh":225,"en":225},"Pichia-CLM 从蛋白序列生成密码子优化 CDS 的模型架构图","\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fc1d9d40f0377f0580657aee4e",{"role":194,"title":228,"caption":230,"alt":232,"position":44,"imageUrl":234},{"zh":229,"en":229},"同一套指标下的结果对照",{"zh":231,"en":231},"Jupyter 中的脚本把模型生成序列和规则优化序列放进同一张表，逐项比较 GC、CAI、ENC 与稀有密码子。31.92 对 76.04 的综合评分，也由此成为下一轮融合两条路线的起点。",{"zh":233,"en":233},"Jupyter 中的密码子优化指标计算与对照表","\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fcc7dde33da92ca492db760e16",{"role":194,"alt":236,"position":44,"imageUrl":238},{"zh":237,"en":237},"TJU-Pichia计算与数据分析记录","\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fcf97ec1c379efccc876593d7c",{"id":240,"eyebrow":241,"title":243,"theme":245,"type":246,"highlight":247,"full":256},"team-interview",{"zh":242,"en":242},"队伍专访",{"zh":244,"en":244},"听 TJU-Pichia 讲述项目路径","media","interview",{"version":248,"title":249,"description":251,"durationSeconds":253,"videoUrl":254,"posterUrl":255},"highlight",{"zh":250,"en":250},"项目精华",{"zh":252,"en":252},"项目核心内容与队伍表达的精华剪辑。",121.1,"\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fcfd7d6a30f121895973c9d04f","\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fc1c17437e697af64355b47428",{"version":257,"title":258,"description":260,"durationSeconds":262,"videoUrl":263,"posterUrl":255},"full",{"zh":259,"en":259},"完整访谈",{"zh":261,"en":261},"按采访问题完整收录的队伍访谈。",379.266666,"\u002Fapi\u002Fteam-showcase\u002Ftju-pichia\u002Fmedia\u002Fcb7bb58b0e643e0ca6bbb2e78"]