Identification of effective diagnostic biomarker and immune cell infiltration characteristics in acute liver failure by integrating bioinformatics analysis and machine-learning strategies

Background: To determine effective biomarkers for the diagnosis of acute liver failure (ALF) and explore the characteristics of the immune cell infiltration of ALF. Methods: We analyzed the differentially expressed genes (DEGs) between ALF and control samples in GSE38941, GSE62029, GSE96851, GSE120652, and merged datasets. Co-expressed DEGs (co-DEGs) identified from the five datasets were analyzed for enrichment analysis. We further constructed a PPI network of co-DEGs using the STRING database. Then, we integrated the two kinds of machine-learning strategies to identify diagnostic biomarkers of top hub genes screened based on MCC and Degree methods. And the potential diagnostic performance of the biomarkers for ALF was estimated using the AUC values. Data from GSE14668, GSE74000, and GSE96851 databases was performed as external verification sets to validate the expression level of potential diagnostic biomarkers. Furthermore, we analyzed the difference in the protein level of diagnostic biomarkers between normal and ALF mice models. Finally, we used CIBERSORT to estimate relative infiltration levels of 22 immune cell subsets in ALF samples and further analyzed the relationships between the diagnostic biomarkers and infiltrated immune cells. Results: A total of 200 co-DEGs were screened. Enrichment analyses depicted that they are highly enriched in metabolism and matrix collagen production-associated processes. The top 28 hub genes were obtained by integrating MCC and Degree methods. Then, the collagen type IV alpha 2 chain (COL4A2) was regarded as the diagnostic biomarker and showed excellent specificity and sensitivity. COL4A2 also showed a statistically significant difference and excellent diagnostic effectiveness in the verification set. In addition, there was a significant upregulation in the COL4A2 protein level in ALF mice models compared with the normal group. CIBERSORT analysis showed that activated CD4 T cells, plasma cells, macrophages, and monocytes may be implicated in the progress of ALF. In addition, COL4A2 showed different degrees of correlation with immune cells. Conclusion: In conclusion, COL4A2 may be a diagnostic biomarker for ALF, and immune cell infiltration may have important implications for the occurrence and progression of ALF.