研究设计:新冠肺炎人工智能导航。第0步:使用名为布尔等价相关聚类(BECC133)的机器学习工具挖掘超过45000个人类、小鼠和大鼠基因表达数据库,以确定宿主对病毒大流行(ViP)的不变反应。在COVID-19大流行开始时,由于缺乏足够多的COVID-19数据集,这些ViP签名仅对来自过去大流行的两个数据集(流感和禽流感;GSE47963, n = 438;GSE113211, n = 118),并在未经进一步培训的情况下用于前瞻性分析来自当前大流行(即COVID-19;N =来自不同数据集的727个样本)。一个由20个基因分类的疾病严重程度的子集称为严重- vip (sViP)特征。ViP特征似乎捕捉到了“不变的”宿主反应,即包括COVID-19在内的所有病毒大流行诱导的宿主免疫反应的共同基本性质。Step1:在代表大量肺部疾病的不同转录组数据集上分析ViP/sViP签名集和cov -肺特异性13基因签名集;这些努力确定COVID-19肺部疾病最接近特发性肺纤维化(IPF); both conditions induced a common array of gene signatures. Step 2: Clinically useful whole-blood and PBMC-derived prognostic signatures previously validated in IPF27 showed crossover efficacy in COVID-19, and vice versa. Step 3: Gene signatures of alveolar type II (AT2) cytopathic changes that are known to fuel IPF were analyzed in COVID-19 lung, and predicted shared features were validated in human and hamster lungs and lung-organoid derived models. Step 4: Protein- protein interaction (PPI) network built using sViP and AT2 cytopathy-related signatures was analyzed to pinpoint ER stress as a major shared feature in COVID-19 lung disease and IPF, which was subsequently validated in human and hamster lungs. Credit:eBioMedicine(2022)。DOI: 10.1016 / j.ebiom.2022.104185