Deep learning predicts potential reassortments of avian H5N1 with human influenza viruses

作者：Junqing Wei, Sen Zhang, Yingde Li, Shuyang Jiang, Si-Rui Yan, Ying Liu, Yuehong Chen, Ye Feng, Xiao Ding, Yuchang Li, Xiaoping Kang, Wei Liu, Aiping Wu, Tao Jiang, Yigang Tong, Jing Li · 发表于：National Science Review · 年份：2025 · DOI：10.1093/nsr/nwaf396 · 被引用次数：3 · 研究领域：Influenza Virus Research Studies

ABSTRACT Frequent infection cases with avian H5N1 influenza A viruses (IAVs) are posing pandemic risks of human/avian-reassorted IAVs. We aimed to build an attentional deep learning model named HAIRANGE, for predicting potentially human-adapted reassortment of H5N1 and human IAVs. A biologically relevant and non-pretrained embedder named Codon2Vec in HAIRANGE performed competitively in benchmarking against other embedders, such as ESM2, DNABERT2 and others, indicating a high association of genomic context with viral hosts or serotypes, for IAV RNA polymerase-related genes. HAIRANGE accurately predicted the adaptation of each polymerase-related gene and the adaptive polymerase-related gene reassortment with polymerase activity validated by in vitro reporting assay. Worryingly, an adaptive reassortment between avian H5N1 and human H3N2 IAVs was predicted by HAIRANGE and validated by polymerase activity assay. Summarily, HAIRANGE can predict adaptive IAV reassortment based on embedded genomic context. Current avian H5N1 IAV is posing pandemic potential via possible reassortment with human IAVs.