#Please insert up references in the next lines (line starts with keyword UP)
UP	arb.hlp
UP	glossary.hlp


TITLE		Estimation of Bootstrap by Parsimony

OCCURRENCE	ARB_NT/Tree/Parsimony

DESCRIPTION	Given a large tree, traditional ways to calculate bootstrap
		values are by magnitudes to slow. So a faster algorithm was
		developed:

                        the bootstrap value for each branch is calculated under
                        the assumption that all other branches have a 100% value.
                        Doing this we get an upper limit for the real bootstrap values.


NOTES		The program does not use the traditional Monte Carlo method to
		estimate the bootstrap values, but calculates them correctly
		under the assumption that the tree changes only locally.
		Try different filters and see the effect on the tree.

SECTION         ALGORITHM

                For each branch B do:

		      a                 b
		       \               /
			>-------------<
		       /	B      \
                      c                 d

		exchange a with b ( or a with d ) and count all coloums in the alignment
		with a greater/smaller/equal minimal number of mutations
		than the original tree.

		result:		n_plus, n_minus, n_equal
				freq_n_plus = n_plus/ (seq_len)
				...

		Bootstrap value = sum of

			for all i = 1.. seqlen do
				for all combinations of np, nm,ne with np - nm == i do
					sum +=	freq_n_plus  ^ np *
						freq_n_minus ^ nm *
						freq_n_equal ^ ne *
						seq_len! / np! /nm! /ne!
				done
			done

SECTION         PUBLIC

                This algorithm is not published and I am not going to publish
		it. If you feel the strong need to do this, please don't forget
		to mention me (Oliver Strunk <strunk@mikro.biologie.tu-muenchen.de>)

WARNINGS	Use filters to exclude superfluous gaps and to
		increase bootstrap values

BUGS		Does not work with weights
		Does not work with proteins