Scholarlink

Background <p>COVID-19 exhibits seasonal epidemics with high risk of mortality in vulnerable population. Identifying accurate parameters for predicting severity and adverse outcomes remains of great clinical significance.</p> Methods <p>An in-depth serum proteomics was performed on 20 samples from healthy controls (HC) and 108 samples from COVID-19 patients with mild or severe symptoms. Machine learning algorithms were integrated to identify signature proteins, which were further validated via ELISA across the discovery cohort, an external COVID-19 cohort (<i>N</i> = 48) , as well as an influenza A cohort (<i>N</i> = 104). Using a Gradient Boosting model, we developed a Severity-Resistance Index (SRI) to predict the severe risk. Furthermore, we applied unsupervised multi-modal clustering to integrate proteomics with electronic health records (EHR) data for the outcome prediction of severe COVID-19.</p> Results <p>Four proteins including cellular communication network factor 1 (CCN1), selenium-binding protein 1 (SELENBP1), phospholipase A2 group IIA (PLA2G2A) and surfactant protein B (SFTPB) were screened out with remarkable changes in severe COVID-19 serum. This four-protein combination could effectively distinguish mild cases from severe cases with an AUC of  0.829. SRI derived from these four proteins accurately predicted disease severity in both COVID-19 (training set: AUC = 0.985; test set: AUC = 0.827) and influenza patients (training AUC = 1.000; test AUC = 0.889). Additionally, unsupervised clustering identified a distinct mortality-associated subgroup (CS3) within COVID-19 cases which could be effectively distinguished from CS1 and CS2 subgroups using SRI and three key EHR indicators including monocyte percentage, alanine aminotransferase and uric acid with the AUC values of 0.930 (CS1 vs CS3) and 0.781 (CS2 vs CS3), respectively.</p> Conclusions <p>Our study leverages an integrated framework combining artificial intelligence, in-depth proteomics and EHRs to develop a sequential risk assessment tool. While the SRI alone identifies patients at risk for severe illness, SRI incorporating with EHR data further pinpoints individuals at high mortality risk. This strategy therefore converts multidimensional omics data into actionable clinical indication, facilitating risk stratification and precision intervention for viral pneumonia.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

In-depth serum proteomics atlas of COVID-19 defines a Severity-Resistance Index from a four-protein panel for disease severity and prognosis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月13日

查看全文
  • Liheng He,
  • Kai Wei,
  • Junyao Lu,
  • Dong Wei,
  • Xiaoxiao Meng,
  • Peng Chen,
  • Lingli Zeng,
  • Ping Ji,
  • Lian Zhang,
  • Xinxin Zhang,
  • Ruilan Wang,
  • Jieming Qu,
  • Jie Xu,
  • Yingying Chen,
  • Bing Su,
  • Liyun Yuan,
  • Ying Wang

摘要

Background

COVID-19 exhibits seasonal epidemics with high risk of mortality in vulnerable population. Identifying accurate parameters for predicting severity and adverse outcomes remains of great clinical significance.

Methods

An in-depth serum proteomics was performed on 20 samples from healthy controls (HC) and 108 samples from COVID-19 patients with mild or severe symptoms. Machine learning algorithms were integrated to identify signature proteins, which were further validated via ELISA across the discovery cohort, an external COVID-19 cohort (N = 48) , as well as an influenza A cohort (N = 104). Using a Gradient Boosting model, we developed a Severity-Resistance Index (SRI) to predict the severe risk. Furthermore, we applied unsupervised multi-modal clustering to integrate proteomics with electronic health records (EHR) data for the outcome prediction of severe COVID-19.

Results

Four proteins including cellular communication network factor 1 (CCN1), selenium-binding protein 1 (SELENBP1), phospholipase A2 group IIA (PLA2G2A) and surfactant protein B (SFTPB) were screened out with remarkable changes in severe COVID-19 serum. This four-protein combination could effectively distinguish mild cases from severe cases with an AUC of  0.829. SRI derived from these four proteins accurately predicted disease severity in both COVID-19 (training set: AUC = 0.985; test set: AUC = 0.827) and influenza patients (training AUC = 1.000; test AUC = 0.889). Additionally, unsupervised clustering identified a distinct mortality-associated subgroup (CS3) within COVID-19 cases which could be effectively distinguished from CS1 and CS2 subgroups using SRI and three key EHR indicators including monocyte percentage, alanine aminotransferase and uric acid with the AUC values of 0.930 (CS1 vs CS3) and 0.781 (CS2 vs CS3), respectively.

Conclusions

Our study leverages an integrated framework combining artificial intelligence, in-depth proteomics and EHRs to develop a sequential risk assessment tool. While the SRI alone identifies patients at risk for severe illness, SRI incorporating with EHR data further pinpoints individuals at high mortality risk. This strategy therefore converts multidimensional omics data into actionable clinical indication, facilitating risk stratification and precision intervention for viral pneumonia.

Graphical Abstract