Identification and validation of key host genes associated with porcine H1N1 infection based on integrated machine learning algorithms.

Objective: Swine H1N1 influenza is a critical zoonotic pathogen threatening pig industry economy and public health. The host molecular regulatory network and core genes of H1N1 infection remain unclear, hindering targeted prevention and therapy. Traditional experimental methods fail to efficiently mine high-dimensional transcriptomic data, making precise screening of infection biomarkers difficult. Methods: Transcriptome data (GSE40092) were analyzed to obtain porcine lung DEGs upon H1N1 infection, followed by GO/KEGG functional enrichment. Four machine learning algorithms (LASSO, random forest, SVM-RFE, XGBoost) coupled with stratified nested 5-fold cross-validation screened core genes. Feature stability analysis and external dataset GSE28871 validated biomarker robustness. A gradient-dose H1N1 piglet model and Western blot verified the key gene's Results: A total of 310 H1N1-related DEGs were enriched in immune, inflammatory and viral signaling pathways. All four models accurately discriminated infected and normal lung samples, with SPP1 as the only shared core gene. Cross-validation proved SPP1 screening free of overfitting; external validation yielded an AUC of 0.889, 83.3% sensitivity and 100% specificity. Conclusion: This study combined transcriptomics and multi-machine learning to identify and verify host genes for swine H1N1 infection. SPP1 acts as a stable diagnostic biomarker whose reduced expression correlates with disease progression. Our results reveal new molecular mechanisms of H1N1 pathogenesis and offer a candidate target for swine flu control and zoonotic risk intervention.