Jackknife and Bootstrap Tests of the Composition Vector Trees

Zuo, Guanghong; Xu, Zhao; Yu, Hongjie; Hao, Bailin

As CV method does not use sequence alignment, statistical re-sampling cannot be carried out in the usual way of random choice of nucleotide or amino acid sites with replacement. Instead, we pick up proteins at random from the pool of all proteins in the genome of an organism. We used four datasets of protein sequences encoded in the genome: On each of these datasets, jackknife and bootstrap tests are performed in the following way. In the CV method, a species is represented by a composition vector made of overlapping K-residues, designated as "K-peptides" hereafter, from all proteins in the genome. To do jackknife tests, we first take randomly 90% of proteins from the whole protein pool. This is done for all species and a CVTree is constructed by carrying out the crucial "subtraction procedure" (1). The topological distance between this tree and the original CVTree inferred from the whole protein pool is calculated. This re-sampling is performed 100 times and the average topological distance between these 100 trees and the original 100% CVTree at the same K is taken. Then the protein fraction is decreased to 80%, 70%, …, 10% and the average topological distance at a given K is plotted against the protein fraction (Figure 1) .