Y to accept these or type the letter for one to change
Most of the input options (L, R, S, J, and M) are as given in the Distance Matrix Programs documentation file, that file, and their input format is the same as given there. The O (Outgroup) option is described in the main documentation file of this package. It is not available when the UPGMA option is selected. The Jumble option (J) does not allow multiple jumbles (as most of the other programs that have it do), as there is no objective way of choosing which of the multiple results is best, there being no explicit criterion for optimality of the tree.
Option N chooses between the Neighbor-Joining and UPGMA methods. Option S is the usual Subreplication option. Here, however, it is present only to allow NEIGHBOR to read the input data: the number of replicates is actually ignored, even though it is read in. Note that this means that one cannot use it to have missing data in the input file, if NEIGHBOR is to be used.
The output consists of an tree (rooted if UPGMA, unrooted if Neighbor-Joining) and the lengths of the interior segments. The Average Percent Standard Deviation is not computed or printed out. If the tree found by Neighbor is fed into FITCH as a User Tree, it will compute this quantity if one also selects the N option of FITCH to ensure that none of the branch lengths is re-estimated.
As NEIGHBOR runs it prints out an account of the successive clustering levels, if you allow it to. This is mostly for reassurance and can be suppressed using menu option 2. In this printout of cluster levels the word "OTU" refers to a tip species, and the word "NODE" to an interior node of the resulting tree.
The constants available for modification at the beginning of the program are "namelength" which gives the length of a species name, and the usual boolean constants that initiliaze the terminal type. There is no feature saving multiply trees tied for best, partly because we do not expect exact ties except in cases where the branch lengths make the nature of the tie obvious, as when a branch is of zero length.
The major advantage of NEIGHBOR is its speed: it requires a time only
proportional to the square of the number of species. It is
significantly faster than version 3.5 of this program. By contrast
FITCH and KITSCH require a time that rises as the fourth power of the
number of species. Thus NEIGHBOR is well-suited to bootstrapping
studies and to analysis of very large trees. Our simulation studies
(Kuhner and Felsenstein, 1994) show that, contrary to statements in the
literature by others, NEIGHBOR does not get as accurate an estimate of
the phylogeny as does FITCH. However it does nearly as well, and in
view of its speed this will make it a quite useful program.
__________________________________________________________________