Neural Replicator Analysis for virus genomes binomial systematics in metagenomics

TL;DR

This paper introduces neural replicator analysis (NRA) as a novel method for classifying virus genomes into binomial categories using only genomic sequences, without alignment or phenotypic data, especially useful for metagenomic data.

Contribution

The paper presents NRA as a new approach for virus genome classification that works without sequence alignment or phenotypic information, applicable across various virus families.

Findings

NRA effectively classifies virus genomes into binomial categories.

The method reveals virus properties and host relationships.

Applicable to diverse virus genome types, including linear and circular DNA and RNA.

Abstract

We have presented some arguments to substantiate the usefulness of neural replicator analysis (NRA) for constructing variants of the natural binomial classification of virus genomes based only on knowledge of their complete genomic sequences, without involving other data on the phenotype, functions, encoded proteins, etc., and also without the need of genomic sequences alignment. Perhaps this will make sense when processing metagenomic data. This makes it possible to construct the binomial classification accepted for the viruses themselves. We restrict ourselves to three families of viruses having dsDNA circular genomes (Papillomaviridae, Polyomaviridae and Caulimoviridae) and partly to the family Geminiviridae having ssDNA genomes though the approach presented can be also applied to genomes of other dsDNA, ssDNA and ssRNA viruses, including linear ones (some results for Mitoviridae are also presented). It is argued that binomial classification of virus genomes which is difficult to apply in all cases can nevertheless be informative tool of revealing virus properties, areal of hosts, forms of diseases and can also show the connections of the viruses belonging to different families and even to different kingdoms.