Hantavirus species detection from viral genome sequences using explainable artificial intelligence techniques

Hantaviruses are important zoonotic pathogens associated with severe human diseases, including hemorrhagic fever with renal syndrome and hantavirus cardiopulmonary syndrome. The World Health Organization (WHO) has recognized the importance of strengthening hantavirus surveillance and early detection because of their public health impact. Nevertheless, Artificial Intelligence (AI)-based approaches for hantavirus species identification from viral genome sequences remain largely unexplored. This study developed and evaluated an explainable AI-based approach for hantavirus species identification using viral genome sequence fragments representing six hantavirus species. A transformer-based genomic representation approach (DNABERT-2) was used to generate representations for various Machine Learning (ML) and Deep Learning (DL) models. Sequence redundancy was quantified before model development, and performance was assessed using a 99%-sequence-identity group-aware partitioning strategy with five-fold group-aware cross-validation and an independent internal test set. Direct sequence-similarity auditing confirmed that no sequence pair with ≥ 99% global nucleotide identity crossed the development and test boundary. LightGBM achieved the highest mean group-aware cross-validation performance, with an Accuracy and weighted F1-Score of 99.51%, Cohen’s Kappa coefficient of 99.41%, and Brier Score of 0.19%. On the independent internal test set, LightGBM achieved an Accuracy and weighted F1-Score of 99.25%. SHAP analysis identified influential nucleotide sequence tokens contributing to LightGBM predictions, while a separate DNABERT-2 attention-based analysis identified highly attended nucleotide sequence patterns within the genomic encoder. These findings demonstrate the potential of the proposed approach as an auxiliary method for post-sequencing hantavirus species identification, genomic surveillance, and sequence-based research.