Our understanding of viral sequence space has significantly expanded due to advancements in sequencing technologies. However, knowledge of archaeal viruses, especially outside extreme environments, remains limited. This paper presents MArVD2, an upgraded machine learning tool using a random forest algorithm, trained on a newly curated dataset of archaeal viruses. MArVD2 shows improved scalability, usability, and flexibility compared to its predecessor. Benchmarking revealed that a model trained on viruses from hypersaline, marine, and hot spring environments correctly classified 85% of archaeal viruses with a false detection rate below 2% using a prediction threshold of 80%.

Dean Vik, Benjamin Bolduc, Simon Roux, Christine L. Sun, Akbar Adjie Pratama, Mart Krupovic, Matthew B. Sullivan