★高(同行评审一手论文) Osinski B.L. 等,Artificial intelligence-augmented histopathologic review using image analysis to optimize DNA yield from formalin-fixed paraffin-embedded slides,Modern Pathology 2022;35(PMID 36198869)。U-Net 细胞分割+多视野 CNN 肿瘤区分割,特征与已知 DNA 产量拟合正则化线性模型 R=0.85;501 张结直肠癌切片内部验证,100~2000 ng 目标区间命中率高 25%,提取尝试次数无统计学差异。https://doi.org/10.1038/s41379-022-01161-0 ;预印本:https://arxiv.org/abs/2203.13948 ;题录:https://europepmc.org/article/MED/36198869
★高(同行评审一手论文) A machine learning-based approach for improving plasmid DNA production in Escherichia coli fed-batch fermentations,Biotechnology Journal 2024(DOI 10.1002/biot.202400140 ,PMID 38896410)。随机森林为最优算法预测升温策略;50 L 反应器质粒产量 1,167.74 mg/L、比产率 8.87 mg/L/OD600,较对照高 71% 与 21%。https://europepmc.org/article/MED/38896410
★高(同行评审验证研究+厂商产品页) Validation of a neural network approach for STR typing to replace human reading,Forensic Science International: Genetics 2021;55(DOI 10.1016/j.fsigen.2021.102591 ,PMID 34530398):商用软件 FaSTR DNA 内嵌 ANN,参考图谱等位基因峰检出准确率 99.7%,南澳法医实验室据此实施"单人+AI"判读流程。产品页:https://www.strmix.com/fastr/ ;题录:https://europepmc.org/article/MED/34530398
★高(厂商技术说明,性能为厂商宣称) Thermo Fisher Scientific Technical Note 52918《Acclaro phenol contaminant ID — Detection of phenol in nucleic acid samples using the NanoDrop One spectrophotometer》(作者 Loughrey、Matlock)。官方关键词含 Chemometrics/Spectral Analysis/Sample Intelligence;Acclaro 可识别蛋白、苯酚、胍盐污染并以黄色图标提示,苯酚提示下限约 18.75 ppm,高污染下校正浓度在纯样 ±10% 内、平均标准差<5 ng/μL;Table 1 给出 TE 缓冲液中 dsDNA 260/280 1.8~2.0、260/230 1.8~2.2,RNA 2.0~2.2、1.8~2.2,纯苯酚 1.6、2,并明确"纯度比值不足以判定污染身份与含量"。https://documents.thermofisher.com/TFS-Assets/LSG/brochures/TN52918-Acclaro-Phenol-Tech-Note-NanoDrop-One.pdf
★高(厂商官方页面,中国) Thermo Fisher 中国官网 NanoDrop 产品页:中文原话"内置的 Acclaro 样本智能污染物识别技术可帮助识别并校正样本中存在的污染物""可选配的 Acclaro Pro 软件,该软件采用先进算法,可对高浓度样本进行高精度定量""用于合规的可选机载 SciVault 2 软件……有助于满足 FDA 21 CFR Part 11 电子记录法规要求,并可通过本地仪器上的 RESTful API 直接与您的 LIMS 环境无缝集成";检出限 0.2 ng/μL dsDNA,光谱范围 190~850 nm。另:NanoDrop One 污染判读应用说明 https://documents.thermofisher.cn/TFS-Assets/MSD/Application-Notes/nanodrop-one-phenol-protein-contaminants-rtqpcr-en-tn53472.pdf 。https://www.thermofisher.cn/cn/zh/home/industrial/spectroscopy-elemental-isotope-analysis/molecular-spectroscopy/uv-vis-spectrophotometry/instruments/nanodrop/instruments/nanodro-one.html
★高(厂商官方页面;具体阈值未确认) Implen 官方"Sample Control Technology for Nucleic Acid and Protein Quantification"技术说明页(Technical Note 10;页面正文因抓取受限,逐项判读波长与阈值记为未确认,未见 ML/AI 表述)与独立可选的 CFR21 合规软件包(审计追踪/权限)。https://implen.com/implen-academy/technical-notes/sample-control-technology/ ;https://www.implen.de/cfr21-software/ ;中文站:https://implen.cn/sample-control-technology-for-nucleic-acid-protein-quantification/
★中(厂商产品目录页) Eppendorf 光度计产品线(BioPhotometer D30 等,双/多波长 260/280、260/230 核酸与蛋白方法,属确定性比值运算;官网未检索到机器学习或 Part 11 表述,操作手册页访问受限)。https://www.eppendorf.com/us-en/Products/Photometry-c-WebPMain-H-44540
★高(厂商官方应用说明/合规页) Molecular Devices:PathCheck 技术将吸光度自动折算为 1 cm 光径(应用说明)、SoftMax Pro GxP 提供 Part 11 类合规;其机器学习产品为 ClonePix 克隆筛选,不用于吸光度定量质控。https://www.moleculardevices.com/en/assets/app-note/br/optical-density-measurements-automatically-corrected-to-1-cm-pathlength-with-pathcheck-technology ;https://www.moleculardevices.com/products/gxp-compliance-solutions/softmax-pro-gxp-software
★高(一手文献;本次经 Semantic Scholar 摘要接口转取,出版商页访问受限) Wilfinger W.W.、Mackey K.R.、Chomczynski P.,Effect of pH and ionic strength on the spectrophotometric assessment of nucleic acid purity,BioTechniques 1997;22:474-481(DOI 10.2144/97223st01):测量用水 pH 由约 5.4 调至 7.5~8.5 可使 RNA A260/280 由约 1.5 升至 2.0;260 nm 吸收峰在 pH 8.5 Na2HPO4 中向短波移 2 nm;建议在 1~3 mM Na2HPO4、pH 8.0~8.5 下测定,碱性条件下蛋白污染更易检出。DOI:https://doi.org/10.2144/97223st01 ;https://www.semanticscholar.org/paper/1120a33627cf3e1ae51868252a21509a4dea745b (出版商页与 DOI 解析页在本次核查环境返回 403,上述结论逐句取自 Semantic Scholar 摘要接口,标注为间接来源)
★高(一手文献) Mach S.A.、Middaugh C.R.、Lewis R.N.,Detection of proteins and phenol in DNA samples with second-derivative absorption spectroscopy,Analytical Biochemistry 1992(DOI 10.1016/0003-2697(92)90270-H ,PMID 1375815):近紫外二阶导数光谱+矩阵最小二乘多元组分分析,可在吸光度 0.1 的 DNA 样品中检出<0.002 吸光度的色氨酸/酪氨酸/苯丙氨酸/苯酚,实际蛋白检出限<0.003;蛋白-DNA 混合物误差<1%、DNA-RNA 混合物约 2%;作者指出可用普通光谱软件在数秒内完成。https://europepmc.org/article/MED/1375815
★高(同行评审一手论文·仪器间比对) Comparison of DeNovix, NanoDrop and Qubit for DNA quantification and impurity detection of bacterial DNA extracts,PLOS ONE 2024(DOI 10.1371/journal.pone.0305650 ,PMID 38885212):同批肺炎链球菌基因组 DNA(Roche 试剂盒提取),紫外法浓度较 Qubit 高 3~4 倍,该比值随 A260/280>2.0 递增;A260/280 与 A260/230 在不同紫外仪器与一次冻融条件下变化可忽略。https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0305650
★中(预印本,未经同行评审) Optimizing RNA yield using deep neural networks coupled to massively parallel screening,arXiv:2608.23722(提交 2026-08-24):10⁵ 条随机寡核苷酸文库、Illumina 平行定量 DNA/RNA 丰度,one-hot 编码+CNN 预测 IVT RNA 产量,留出集 预测 vs 实测 Pearson r=0.94;作者称可部署于生产环境对序列设计按预测产量排序。对象为体外转录产量,非提取回收率。https://arxiv.org/abs/2608.23722
★高(同行评审一手论文·相邻领域) Predicting purification process fit of monoclonal antibodies using machine learning,MAbs 2025;17(1):2439988(DOI 10.1080/19420862.2024.2439988 ,PMID 39782766 ,作者单位 Genentech 纯化/微生物/病毒学部):高通量 batch-binding 数据+PCA+核岭回归的 QSPR 框架,97 株单抗×5 种树脂×两种盐系与 pH/盐浓度区间,基准比较 4 组描述符(结构模型与蛋白语言模型)。具体样本外精度数值本期未能核实(T&F 全文页访问受限,Europe PMC 摘要被截断),记为未确认。https://europepmc.org/articles/PMC11730362 ;题录:https://europepmc.org/article/MED/39782766(DOI 10.1080/19420862.2024.2439988)
★高(同行评审一手论文·开源工具) GelGenie: an AI-powered framework for gel electrophoresis image analysis,Nature Communications 2025(DOI 10.1038/s41467-025-59189-0 ,PMID 40324991):以 500+ 张人工标注胶图训练 U-Net 做条带分割,跨实验室结果与原作者定量结论一致,模型开源、可本地设备运行。https://www.nature.com/articles/s41467-025-59189-0
★高(一手方法学论文) The RIN: an RNA integrity number for assigning integrity values to RNA measurements,BMC Molecular Biology 2006;7:3(DOI 10.1186/1471-2199-7-3 ,PMID 16448564):RIN 由微毛细管电泳信号自动特征选择+贝叶斯学习技术构建回归模型得到(Agilent 2100 大样本集),已成为行业通用完整性口径——证明"模型化质控指标"在核酸领域早已规模化,只是对象为完整性而非回收率/纯度比。https://link.springer.com/article/10.1186/1471-2199-7-3
★中(一手论文·相关性而非预测) Establishing a correlation between RIN and A260/280 along with the multivariate evaluation of factors affecting the quality of RNA in cryopreserved cancer bio-specimen,Cell and Tissue Banking 2019;20(DOI 10.1007/s10561-019-09782-7 ,PMID 31444592):冷冻肿瘤组织 RNA 中约 50% 样本 RIN≥6.9 且 A260/280≤2.04、27% RIN≥5 且≤2.08、23% RIN<5 且>2.08;A260/280≤2.08 者 Ct 17~24 均可接受,支持"比值可预测 RNA 质量"但为相关性证据、非建模。https://europepmc.org/article/MED/31444592
★高(法规原文) 21 CFR Part 11(Electronic Records; Electronic Signatures)eCFR 现行文本(§11.10 含系统验证与审计追踪要求)。https://www.ecfr.gov/current/title-21/chapter-I/subchapter-A/part-11
★高(官方指南原文) PIC/S《Guidance on Data Integrity》PI 041-1(Draft 3):ALCOA+、审计追踪与计算机化系统验证原则。https://picscheme.org/users_uploads/news_news_documents/PI_041_1_Draft_3_Guidance_on_Data_Integrity.pdf
★中(指南镜像;因 fda.gov 对本环境访问受限(反爬 404),此项为间接来源) FDA《Data Integrity and Compliance With Drug CGMP: Questions and Answers》(2018 终稿)镜像页:强调原始数据与元数据须完整保留、可复现,删除/覆盖与"黑箱"式处理构成数据完整性风险。https://www.gmp-compliance.org/guidelines/gmp-guideline/fda-guidance-for-industry-data-integrity-and-compliance-with-cgmp (原页 https://www.fda.gov/regulatory-information/search-fda-guidance-documents/data-integrity-and-compliance-drug-cgmp-questions-and-answers 访问受限)
★高(可复现负结果检索式,检索日 2026-09-03) Europe PMC REST 题录检索口径与结果:TITLE:"extraction yield" +(machine OR learning OR predict* OR optimi*)=16 篇且全为植物/超临界/中药提取,无一篇为核酸提取;TITLE:"DNA yield" +(predict* OR machine learning OR artificial OR AI)=1 篇(即 [2]);TITLE:"nucleic acid extraction" +(machine OR artificial intelligence)=1 篇([33]);TITLE:"cell-free DNA" +TITLE:yield=11 篇,均为工艺/临床得率、无预测建模;TITLE:RIN + predict*=0;ABSTRACT:"Bayesian optimization"+"library preparation"=0;TITLE:"A260/280"=2 篇(其中仅 [26] 与本题相关)。可复现检索式(Europe PMC,字段检索语法,均于 2026-09-03 执行):①TITLE:"extraction yield" AND (TITLE:machine OR TITLE:learning OR TITLE:predict* OR TITLE:optimi*)=16;②TITLE:"DNA yield" AND (TITLE:predict* OR TITLE:"machine learning" OR TITLE:artificial OR TITLE:AI)=1;③TITLE:"nucleic acid extraction" AND (TITLE:machine OR TITLE:"artificial intelligence")=1;④TITLE:"cell-free DNA" AND TITLE:yield=11;⑤TITLE:RIN AND (TITLE:predict* OR TITLE:"machine learning")=0;⑥ABSTRACT:"Bayesian optimization" AND ABSTRACT:"library preparation"=0;⑦TITLE:"A260/280" OR TITLE:"260/280"=2;⑧TITLE:"nucleic acid" AND TITLE:quantification AND (TITLE:"machine learning" OR TITLE:neural OR TITLE:predict*)=2。示例 UI 链接(可逐条替换检索式复现):https://europepmc.org/search?query=TITLE%3A%22nucleic%20acid%20extraction%22%20AND%20(TITLE%3Amachine%20OR%20TITLE%3A%22artificial%20intelligence%22) ;REST 端点(偶发 503 限流,建议重试或改用上述 UI):https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=TITLE:%22extraction%20yield%22%20AND%20(TITLE:machine%20OR%20TITLE:learning%20OR%20TITLE:predict*%20OR%20TITLE:optimi*)&format=json
★高(同行评审一手论文) A machine learning-driven robotic system for autonomous nucleic acid extraction and library preparation(iRoNAEaLP),SLAS Technology 2025(DOI 10.1016/j.slast.2025.100370 ,PMID 41274605):LSTM+机器学习自动生成流程、试剂与耗材用量并对接模块动作,覆盖痕量核酸提取、质粒/构建体提取与单细胞/空转 mRNA 建库;其 ML 用于流程生成而非产量或纯度预测(后者与 2026-09-01《无代码 SOP 编排与流程仿真工具》主题相邻)。https://europepmc.org/article/MED/41274605
无纸化记录(ELN/LES/LIMS)在制药与生物技术GxP环境已规模化:全球ELN市场2025年约7.2亿美元、2030年预计10.3亿美元(CAGR 7.3%),主要玩家为Revvity、达索、Benchling、Thermo、Waters、IDBS;中国以本土厂商随GxP项目落地为主——湖北省药品监督检验研究院ELN/LIMS功能扩展合同94.64万元(2025-06)、上海某机构鹰谷ELN+CMS买断预算189万元(2026-06)。数据完整性监管框架(ALCOA+、21 CFR Par