Fig. 1
The workflow of this study. This study presents a comprehensive procedure for analyzing host transcriptional profiles in severe respiratory infections. BALF specimens from 180 critically ill patients were subjected to metatranscriptomic analysis. Samples were categorized into four groups based on microbiological findings: microbial infection, bacterial infection, viral infection, and non-infected. Each group was randomly divided into training and testing sets at a 7:3 ratio. Due to limited fungal samples and substantial interindividual variability, fungal specimens were excluded from subsequent analysis. Differential analysis of host transcriptomes identified differentially expressed genes. SVM, RF, and Lasso methods were utilized for host biomarker selection. Integration learning was employed to construct a final diagnostic model by combining 13 machine learning sub-models. The effectiveness and validation of the model were assessed using five-fold cross-validation on the training set and validation on the testing set.
