SEQBOOT -- Bootstrap, Jackknife, or Permutation Resampling of Molecular Sequence, Restriction Site, Gene Frequency or Character Data

(c) Copyright 1991-1993 by the University of Washington and by Joseph Felsenstein. Written by Joseph Felsenstein. Permission is granted to copy this document provided that no fee is charged for it and that this copyright notice is not removed.

SEQBOOT is a general boostrapping tool. It is intended to allow you to generate multiple data sets that are resampled versions of the input data set. Since almost all programs in the package can analyze these multiple data sets, this allows almost anything in this package to be bootstrapped, jackknifed, or permuted. SEQBOOT can handle molecular sequences, binary characters, restriction sites, or gene frequencies.

To carry out a bootstrap (or jackknife, or permutation test) with some method in the package, you may need to use three programs. First, you need to run SEQBOOT to take the original data set and produce a large number (say 100) of bootstrapped data sets. Then you need to find the phylogeny estimate for each of these, using the particular method of interest. For example, if you were using DNAPARS you would first run SEQBOOT and make a file with 100 bootstrapped data sets. Then you would give this file the proper name to have it be the input file for DNAPARS. Running DNAPARS with the M (Multiple Data Sets) menu choice and informaing it to expect 100 data sets, you would generate a big output file as well as a treefile with the trees from the 100 data sets. This treefile could be renamed so that it would serve as the input for CONSENSE. When CONSENSE is run the majority rule consensus tree will result, showing the outcome of the analysis.

This may sound tedious, but the run of CONSENSE is fast, and that of SEQBOOT is fairly fast, so that it will not actually take any longer than a run of a single bootstrap program with the same original data and the same number of replicates. It is not very hard and allows bootstrapping on many of the methods in this package. The same steps are necessary with all of them. Doing things this way some of the intermediate files (the tree file from the DNAPARS run, for example) can be used to summarize the results of the bootstrap in other ways than the majority rule consensus method does.

If you are using the Distance Matrix programs, you will have to add one extra step to this, calculating distance matrices from each of the replicate data sets, using DNADIST or GENDIST. So (for example) you would run SEQBOOT, then run DNADIST using the output of SEQBOOT as its input, then run (say) NEIGHBOR using the output of DNADIST as its input, and then run CONSENSE using the tree file from NEIGHBOR as its input.

The resampling methods available are three:

1. . The bootstrap. Bootstrapping was invented by Bradley Efron in 1979, and its use in phylogeny estimation was introduced by me (Felsenstein, 1985b). It involves creating a new data set by sampling N characters randomly with replacement, so that the resulting data set has the same size as the original, but some characters have been left out and others are duplicated. The random variation of the results from analyzing these bootstrapped data sets can be shown statistically to be typical of the variation that you would get from collecting new data sets. The method assumes that the characters evolve independently, an assumption that may not be realistic for many kinds of data.

2. . Delete-half-jackknifing. This alternative to the bootstrap involves sampling a random half of the characters, and including them in the data but dropping the others. The resulting data sets are half the size of the original, and no characters are duplicated. The random variation from doing this should be very similar to that obtained from the bootstrap. The method is advocated by Wu (1986).

3. . Permuting species within characters. This method of resampling (well, OK, it may not be best to call it resampling) was introduced by Archie (1989) and Faith (1990; see also Faith and Cranston, 1991). It involves permuting the columns of the data matrix separately. This produces data matrices that have the same number and kinds of characters but no taxonomic structure. It is used for different purposes than the bootstrap, as it tests not the variation around an estimated tree but the hypothesis that there is no taxonomic structure in the data: if a statistic such as number of steps is significantly smaller in the actual data than it is in replicates that are permuted, then we can argue that there is some taxonomic structure in the data (though perhaps it might be just a pair of sibling species).

The data input file is of standard form for molecular sequences (either in interleaved or sequential form), restriction sites, gene frequencies, or binary morphological characters. The only options that can be present in the input file are W (Weights) and F (Factors), the latter only in the case of binary (0,1) characters. The Weights can only be 0 or 1, and act to select the characters (or sites) that will be used in the resampling, the others being ignored and always omitted from the output data sets. The Factors option allows us to specify that groups of binary characters represent one multistate character. When sampling is done they will be sampled or omitted together, and when permutations of species are done they will all have the same permutation, as would happen if they really were just one column in the data matrix. For futher description of the F (Factors) option see the Discrete Characters Programs documentation file.

When the program runs it first asks you for a random number seed. This should be an integer greater than zero (and probably less than 32767) and which is of the form 4n+1, that is, it leaves a remainder of 1 when divided by 4. This can be judged by looking at the last two digits of the integer (for instance 7651 is not of form 4n+1 as 51, when divided by 4, leaves the remainder 3).

Then the program shows you a menu to allow you to choose options. The menu looks like this:

Bootstrapped sequences algorithm, version 3.5c

Settings for this run:
  D   Sequence, Morph, Rest., Gene Freqs?  Molecular sequences
  J     Bootstrap, Jackknife, or Permute?  Bootstrap
  R                  How many replicates?  100
  I          Input sequences interleaved?  Yes
  0   Terminal type (IBM PC, VT52, ANSI)?  ANSI
  1    Print out the data at start of run  No
  2  Print indications of progress of run  Yes

Are these settings correct? (type Y or the letter for one to change)

It is important to select the correct data type (the D selection). Each time D is typed the program will change data type, proceeding successively through Molecular Sequences, Discrete Morphological Characters, Restriction Sites, and Gene Frequencies. Some of these will cause additional entries to appear in the menu. If Molecular Sequences or Restriction Sites settintgs and chosen the I (Interleaved) option appears in the menu (and as Molecular Sequences are also the default, it appears in the first menu). It is the usual I option discussed in the Molecular Sequences document file and in the main documentation files for the package.

If the Restriction Sites option is chosen the menu option E appears, which asks whether the input file contains a third number on the first line of the file, for the number of restriction enzymes used to detect these sites. This is necessary because data sets for RESTML need this third number, but other programs do not, and SEQBOOT needs to know what to expect.

If the Gene Frequencies option is chosen an menu option A appears which allows the user to specify that all alleles at each locus are in the input file. The default setting is that one allele is absent at each locus.

The J option allows the user to select Bootstrapping, Delete-Half- Jackknifing, or the Archie-Faith permutation of species within characters. It changes successively among these three each time J is typed.

The R option allows the user to set the number of replicate data sets. This defaults to 100. Most statisticians would be happiest with 1000 to 10,000 replicates in a bootstrap, but 100 gives a good rough picture. You will have to decide this based on how long a running time you want.

Input File

There are relatively few options specified in the input file. The Weights option is allowed (in all cases). So is the Factors option for Discrete Morphological Characters. Other options are not allowed. This is a serious limitation of the program, as users may want to pass other options on to the output data files, for use in the programs. In future versions I hope to gradually add some of the options, particulary the A (Ancestors) option for discrete morphological characters. For the moment if you want to put any such options in you would have to edit them into the output by hand, which will be difficult since the identities of the characters in different columns of the output file will vary as a result of the bootstrapping or jackknifing process.

Output

The numerical options 1 and 2 in the menu also affect the output file. If 1 is chosen (it is off by default) the program will print the original input data set on the output file before the resampled data sets. I cannot actually see why anyone would want to do this. Option 2 toggles the feature (on by default) that prints out up to 20 times during the resampling process notification that the program has completed a certain number of data sets. Thus if 100 resampled data sets are being produced, every 5 data sets a line is printed saying which data set has just been completed. This option should be turned off if the program is running in background and silence is desirable. At the end of execution the program will always (whatever the setting of option 2) print a couple of lines saying that output has been written to the output file.

Size and Speed

Future

------------------ TEST DATA SET --------------------------

    5    6
Alpha     AACAAC
Beta      AACCCC
Gamma     ACCAAC
Delta     CCACCA
Epsilon   CCAAAC

--- CONTENTS OF OUTPUT FILE IF REPLICATES ARE SET TO 10 AND SEED TO 4331 ---
    5    6
Alpha        ACCCAC
Beta         ACCCCC
Gamma        CCCCAC
Delta        CAAACA
Epsilon      CAAAAC
    5    6
Alpha        AAAACC
Beta         AACCCC
Gamma        ACAACC
Delta        CCCCAA
Epsilon      CCAACC
    5    6
Alpha        AAAAAC
Beta         AACCCC
Gamma        CCAAAC


Delta        CCCCCA
Epsilon      CCAAAC
    5    6
Alpha        AAAAAA
Beta         AACCCC
Gamma        ACAAAA
Delta        CCCCCC
Epsilon      CCAAAA
    5    6
Alpha        ACCCAA
Beta         ACCCCC
Gamma        CCCCAA
Delta        CAAACC
Epsilon      CAAAAA
    5    6
Alpha        AAACCC
Beta         AAACCC
Gamma        AACCCC
Delta        CCCAAA
Epsilon      CCCACC
    5    6
Alpha        AACAAC
Beta         AACCCC
Gamma        ACCAAC
Delta        CCACCA
Epsilon      CCAAAC
    5    6
Alpha        ACCCAA
Beta         ACCCCC
Gamma        ACCCAA
Delta        CAAACC
Epsilon      CAAAAA
    5    6
Alpha        AACACC
Beta         AACCCC
Gamma        ACCACC
Delta        CCACAA
Epsilon      CCAACC
    5    6
Alpha        AAAACA
Beta         AAAACC
Gamma        AAAACA
Delta        CCCCAC
Epsilon      CCCCAA