Viral host prediction with Deep learning

This article provided an interesting read on the use of DNN to predict viral host species. Aside from being useful in the development of an index of viral hosts, a pipeline such as that constructed by the authors is essential in the response to a viral outbreak. By characterizing the hosts of a virus during an outbreak, treatment and isolation of the species to prevent zoonotic transfer to humans can be more efficient. The approach taken by this paper involves constructing data input processing methods and architectures for host-virus prediction. Results from the research demonstrated a high success when the LSTM (long short-term memory) architecture was used in combination with CNN or LSTM in isolation (depending on the viral species). Additionally, almost random segmentation of the long viral nucleotide sequences into shorter sequences resulted in more efficient and robust host prediction. When compared to previous, similar studies such as that by Le et al. , it was found that a similar level of prediction was attained without the need for full characterization of host genomes in addition to viral genomes.