Identification of cuproptosis-related molecular subtypes and a novel predictive model of COVID-19 based on machine learning.

Background: To explicate the pathogenic mechanisms of cuproptosis, a newly observed copper induced cell death pattern, in Coronavirus disease 2019 (COVID-19). Methods: Cuproptosis-related subtypes were distinguished in COVID-19 patients and associations between subtypes and immune microenvironment were probed. Three machine algorithms, including LASSO, random forest, and support vector machine, were employed to identify differentially expressed genes between subtypes, which were subsequently used for constructing cuproptosis-related risk score model in the GSE157103 cohort to predict the occurrence of COVID-19. The predictive values of the cuproptosis-related risk score were verified in the GSE163151 cohort, GSE152418 cohort and GSE171110 cohort. A nomogram was created to facilitate the clinical use of this risk score, and its validity was validated through a calibration plot. Finally, the model genes were validated using lung proteomics data from COVID-19 cases and single-cell data. Results: Patients with COVID-19 had higher significantly cuproptosis level in blood leukocytes compared to patients without COVID-19. Two cuproptosis clusters were identified by unsupervised clustering approach and cuproptosis cluster A characterized by T cell receptor signaling pathway had a better prognosis than cuproptosis cluster B. We constructed a cuproptosis-related risk score, based on PDHA1, PDHB, MTF1 and CDKN2A, and a nomogram was created, which both showed excellent predictive values for COVID-19. And the results of proteomics showed that the expression levels of PDHA1 and PDHB were significantly increased in COVID-19 patient samples. Conclusion: Our study constructed and validated an cuproptosis-associated risk model and the risk score can be used as a powerful biomarker for predicting the existence of SARS-CoV-2 infection.