Identification of viruses with the potential to infect human

Zheng Zhang; Zena Cai; Zhiying Tan; Congyu Lu; Gaihua Zhang; Yousong Peng

Two kinds of methods, the sequence alignment-based and alignment-free methods, have been developed to predict the host for viruses. For example, several methods based on k-mers, which were extracted from viral genomes (Ahlgren, Ren et al. 2016, . CC-BY-NC-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/597963 doi: bioRxiv preprint Li and Sun 2018) , and several methods based on sequence blast, have been developed to predict the host of the phage (Bolotin, Quinquis et al. 2005 , Edwards, McNair et al. 2015 . Some studies have also attempted to identify the human virus with these methods. For example, Xu et al. (Xu, Tan et al. 2017) developed SVM models to predict the host of influenza viruses based on word vectors. However, all these studies have focused some type of viruses, such as coronavirus and influenza virus. The methods they developed were not suitable for identification of novel human-infecting viruses from the viral metagenomic sequences. Here, we attempted to build machine learning models for rapid identification of viruses with the potential to infect human.