Constructing a diagnostic prediction model to estimate the severe respiratory syncytial virus pneumonia in children based on machine learning.

Abstract

Severe respiratory syncytial virus (RSV) pneumonia is a leading cause of hospitalization and morbidity in infants and young children. Early identification of severe RSV pneumonia is crucial for timely and effective treatment by pediatricians. Currently, no prediction model exists for identifying severe RSV pneumonia in children. This study aimed to construct a diagnostic prediction model for severe RSV pneumonia in children using a machine learning algorithm. We analyzed data from the Gene Expression Omnibus (GEO) Series, including training dataset GSE246622 and testing dataset GSE105450, to identify differential genes between severe and mild-to-moderate RSV pneumonia in children. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed on the differential genes, followed by the construction of a protein-protein interaction (PPI) network. An artificial neural network (ANN) algorithm was then used to develop and validate a diagnostic prediction model for severe RSV pneumonia in children. We identified 34 differentially expressed genes between the severe and mild-to-moderate RSV pneumonia groups. Enrichment analysis revealed that these genes were primarily related to pathogenic infection and immune response. From the PPI network, we identified 10 hub genes and, using the random forest algorithm, screened out 20 specific genes. The ANN-based diagnostic prediction model achieved an area under the curve (AUC) value of 0.970 in the training group and 0.833 in the testing group, demonstrating the model's accuracy. This study identified specific biomarkers and developed a diagnostic model for severe RSV pneumonia in children. These findings provide a robust foundation for early identification and treatment of severe RSV pneumonia, offering new insights into its pathogenesis and improving pediatric care.