FOKHic: A Framework of \({\varvec{k}}\) -mer Based Hierarchical Classification

The significance of viruses in various ecosystems necessitates a comprehensive examination of viral genomes across diverse datasets, underscoring the growing importance of automated and reliable identification of viruses in metagenomic data. However, the complex composition and large data volumes pose a crucial challenge to taxonomic analysis, particularly in deep hierarchies. In this paper, we propose a new hierarchical classification model, FOKHic, which allows hierarchical classification of viruses from kingdom to family. By combining \(k\) -mer frequency-based coding, principal component analysis, and attention fusion, FOKHic is capable of accomplishing rapid classification of metagenomic viral sequences. Benchmarked on the ICTV virus database, FOKHic exhibits overwhelming performance in terms of accuracy, recall, precision, and F1-score compared to currently popular virus-oriented tools. FOKHic is available at https://github.com/xiaozhangzhang123/FOKHic .