ABSTRACT
Viruses are the most abundant infectious agents on earth, and they infect living organisms such as bacteria, plants and animals, among others. They play an important role in the balance of different ecosystems by modulating microbial populations. In humans, they are responsible for some common diseases and may cause severe illnesses. Viral metagenomic studies have become essential and offer the possibility to understand and extend the knowledge of virus diversity and functionality. For these approaches, an essential step is the classification of viral sequences. In this work, 11 taxonomic classification tools were compared by analysing their performances, in terms of sensitivity and precision, to classify reads at the species and family levels using the same (viral and nonviral) datasets and evaluation metrics, as well as their processing times and memory requirements. The results showed that factors such as richness (numbers of viral species in samples), taxonomic level in the classification and read length influence tool performance. High values of viral richness in samples decreased the performances of most tools. Additionally, the classifications were better at higher taxonomic levels, such as families, compared to lower taxonomic levels, such as species, and were more evident in short reads. The results also indicated that BLAST and Kraken2 were the best tools for classifying all types of reads, while FastViromeExplorer and VirusFinder were only good when used for long reads and Centrifuge, DIAMOND, and One Codex when used for short reads. Regarding nonviral datasets (human and bacterial), all tools correctly classified them as nonviral.
