LEAST-SQUARES ESTIMATION OF MOLECULAR DISTANCE - NOISE ABATEMENT IN PHYLOGENETIC RECONSTRUCTION

Zuckerkandl and Pauling (1962, ''Horizons in Biochemistry,'' pp. 189-2 25, Academic Press, New York) first noticed that the degree of sequenc e similarity between the proteins of different species could be used t o estimate their phylogenetic relationship. Since then models have bee n developed to improve the accuracy of phylogenetic inferences based o n amino acid or DNA sequences. Most of these models were designed to y ield distance measures that are linear with time, on average. The reli ability of phylogenetic reconstruction, however, depends on the varian ce of the distance measure in addition to its expectation. In this pap er we show how the method of generalized least squares can be used to combine data types, each most informative at different points in time, into a single distance measure. This measure reconstructs phylogenies more accurately than existing non-likelihood distance measures. We il lustrate the approach for a two-rate mutation model and demonstrate th at its application provides more accurate phylogenetic reconstruction than do currently available analytical distance measures. (C) 1994 Aca demic Press, Inc.