NEIGHBOR JOINING AND MAXIMUM-LIKELIHOOD WITH RNA SEQUENCES - ADDRESSING THE INTERDEPENDENCE OF SITES

Intrastrand base pairings give ribosomal and other RNA molecules characteristic structures that are important for their function. In order to maintain these structures, a substitution at one paired site may have to be compensated for by an appropriate substitution at the complementary site. Thus paired sites do not evolve independently of one another. Most current methods for inferring phylogeny from molecular sequences assume that the sites are independent and will therefore give statistically unreliable and possibly erroneous results when used on structured RNA sequences. We analyze a new probabilistic model for the evolution of double-stranded RNA molecules that considers substitutions of the base pairs rather than of each of the bases independently. The new model, called the double-stranded model, was incorporated into the neighbor-joining distance and maximum likelihood methods. Computer simulations show that maximum likelihood is very robust to the violation of the assumption of the independence of sites. In contrast, the neighbor-joining method is sensitive to such violations: the double-stranded model can provide a significant increase in the chance of obtaining the correct tree topologies with neighbor joining when distances are large and the tree is difficult to obtain. The new model also leads to lower but more realistic estimates for the statistical confidence in the branch lengths and tree topologies.