Optimally recovering rate variation information from genomes and sequences: pattern filtering.

Abstract

Nucleotide substitution rates vary at different positions within genes and genomes, but rates are difficult to estimate, because they are masked by the stochastic nature of substitutions. In this paper, a linear method, pattern filtering, is described which can optimally separate the signals (related to substitution rates or to other measures of sequence change) from stochastic noise. Pattern filtering promises to be useful in both genomic and molecular evolution studies. In an example using mitochondrial genomes, it is shown that pattern filtering can reveal coding and non-coding regions without the need for prior identification of reading frames or other knowledge of the sequence and promises to be an important tool for genomic analysis. In a second example, it is shown that pattern filtering allows one to classify sites on the basis of an estimator of substitution rates. Using elongation factor EF-1 alpha sequences, it is shown that the fastest sites favor archaea as the sister taxon of eukaryotes, whereas the slower sites support the eocyte prokaryotes as the sister taxon of eukaryotes, suggesting that the former result is an artifact of "long branch attraction."