Machine learning-driven identification of immune signatures from RNA-Seq data in H5N1-infected chickens: a computational Immunology approach

Abstract

H5N1, or highly pathogenic avian influenza, continues to pose a serious risk to public health and poultry. In order to differentiate H5N1-infected from healthy chicken samples and to guide the development of diagnostics and vaccines, we postulated that Machine Learning (ML) applied to RNA sequencing (RNA-Seq) data could detect biologically meaningful immune gene signatures. While ML has been applied to other avian RNA-Seq datasets, to our knowledge, this study represents one of the first applications of interpretable ML to chicken H5N1 RNA-Seq transcriptomic data for immune signature discovery.

RNA-Seq data from H5N1-infected chicken lung and ileum tissues (ArrayExpress E-MTAB-2908) that were publicly available, were chosen. Fivefold stratified cross-validation (~80/20 train/test per fold) was used to train two supervised ML models, such as Random Forest (RF) and linear-kernel Support Vector Machine (SVM). Performance was evaluated using Area Under Curve (AUC) and Receiver Operating Characteristic (ROC) curves.

RF reached a mean AUC = 0.85, while SVM-Linear achieved AUC = 0.75. Top-ranking interferon-stimulated genes (ISGs), including IFIT5, MX1, and OASL, were consistently upregulated in infected samples, indicating activation of type I interferon pathways. Concordant findings across models support the stability and biological relevance of the identified signatures despite the modest sample size.

These genes are highlighted as potential candidates for early molecular diagnostics and for tracking vaccine-induced antiviral immunity. These findings broaden the methodological applications of interpretable ML in avian transcriptomics and computational immunology.