Feature Selection Methods for Identifying Genetic Determinants of Host Species in RNA Viruses

Despite environmental, social and ecological dependencies, emergence of zoonotic viruses in human populations is clearly also affected by genetic factors which determine cross-species transmission potential. RNA viruses pose an interesting case study given their mutation rates are orders of magnitude higher than any other pathogen – as reflected by the recent emergence of SARS and Influenza for example. Here, we show how feature selection techniques can be used to reliably classify viral sequences by host species, and to identify the crucial minority of host-specific sites in pathogen genomic data. The variability in alleles at those sites can be translated into prediction probabilities that a particular pathogen isolate is adapted to a given host. We illustrate the power of these methods by: 1) identifying the sites explaining SARS coronavirus differences between human, bat and palm civet samples; 2) showing how cross species jumps of rabies virus among bat populations can be readily identified; and 3) de novo identification of likely functional influenza host discriminant markers.Author Summary: Moving away from genome scan methods used for human GWAS (ultimately inappropriate for the short highly polymorphic genomes of RNA viruses), our work shows the power and potential of multi-class machine learning algorithms in inferring the functional genetic changes associated with phenotypic change (e.g. crossing a species barrier). We show that even distantly related viruses within a viral family share highly conserved genetic signatures of host specificity; reinforce how fitness landscapes of host adaptation are shaped by host phylogeny; and highlight the evolutionary trajectories of RNA viruses in rapid expansion and under great evolutionary pressure. We do so by (for each dataset) unveiling a set of phenotype characteristic mutations which are shown to be functionally relevant, thus providing new insights into phenotypic relationships between RNA viruses. These methods also provide a solid statistical framework with which the degree of host adaptation can be inferred, thus serving as a valuable tool for studying host transition events with particular relevance for emerging infectious diseases. These methods can then serve as rigorous tools of emergence potential assessment, specifically in scenarios where rapid host classification of newly emerging viruses can be more important than identifying putative functional sites.