RESTML -- Restriction sites Maximum Likelihood program

© Copyright 1986-2000 by The University of Washington. 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.

This program implements a maximum likelihood method for restriction sites data (not restriction fragment data). This program is one of the slowest programs in this package, and can be very tedious to run. It is possible to have the program search for the maximum likelihood tree. It will be more practical for some users (those that do not have fast machines) to use the U (User Tree) option, which takes less run time, optimizing branch lengths and computing likelihoods for particular tree topologies suggested by the user. The model used here is essentially identical to that used by Smouse and Li (1987) who give explicit expressions for computing the likelihood for three-species trees. It does not place prior probabilities on trees as they do. The present program extends their approach to multiple species by a technique which, while it does not give explicit expressions for likelihoods, does enable their computation and the iterative improvement of branch lengths. It also allows for multiple restriction enzymes. The algorithm has been described in a paper (Felsenstein, 1992). Another relevant paper is that of DeBry and Slade (1985).

The assumptions of the present model are:

1. Each restriction site evolves independently.
2. Different lineages evolve independently.
3. Each site undergoes substitution at an expected rate which we specify.
4. Substitutions consist of replacement of a nucleotide by one of the other three nucleotides, chosen at random.

Note that if the existing base is, say, an A, the chance of it being replaced by a G is 1/3, and so is the chance that it is replaced by a T. This means that there can be no difference in the (expected) rate of transitions and transversions. Users who are upset at this might ponder the fact that a version allowing different rates of transitions and transversions would run an estimated 16 times slower. If it also allowed for unequal frequencies of the four bases, it would run about 300,000 times slower! For the moment, until a better method is available, I guess I'll stick with this one!

INPUT FORMAT AND OPTIONS

Subject to these assumptions, the program is an approximately correct maximum likelihood method. The input is fairly standard, with one addition. As usual the first line of the file gives the number of species and the number of sites, but there is also a third number, which is the number of different restriction enzymes that were used to detect the restriction sites. Thus a data set with 10 species and 35 different sites, representing digestion with 4 different enzymes, would have the first line of the data file look like this:

   10   35    4

The first line of the data file will also contain a letter W following these numbers (and separated from them by a space) if the Weights option is being used. As with all programs using the weights option, a line or lines must then follow, before the data, with the weights for each site.

The site data are in standard form. Each species starts with a species name whose maximum length is given by the constant "nmlngth" (whose value in the program as distributed is 10 characters). The name should, as usual, be padded out to that length with blanks if necessary. The sites data then follows, one character per site (any blanks will be skipped and ignored). Like the DNA and protein sequence data, the restriction sites data may be either in the "interleaved" form or the "sequential" form. Note that if you are analyzing restriction sites data with the programs DOLLOP or MIX or other discrete character programs, at the moment those programs do not use the "aligned" or "interleaved" data format. Therefore you may want to avoid that format when you have restriction sites data that you will want to feed into those programs.

The presence of a site is indicated by a "+" and the absence by a "-". I have also allowed the use of "1" and "0" as synonyms for "+" and "-", for compatibility with MIX and DOLLOP which do not allow "+" and "-". If the presence of the site is unknown (for example, if the DNA containing it has been deleted so that one does not know whether it would have contained the site) then the state "?" can be used to indicate that the state of this site is unknown.

User-defined trees may follow the data in the usual way. The trees must be unrooted, which means that at their base they must have a trifurcation.

The options are selected by a menu, which looks like this:

Restriction site Maximum Likelihood method, version 3.6 Settings for this run: U Search for best tree? Yes A Are all sites detected? No S Speedier but rougher analysis? Yes G Global rearrangements? No J Randomize input order of sequences? No. Use input order L Site length? 6 O Outgroup root? No, use as outgroup species 1 M Analyze multiple data sets? No I Input sequences interleaved? Yes 0 Terminal type (IBM PC, ANSI, none)? (none) 1 Print out the data at start of run No 2 Print indications of progress of run Yes 3 Print out tree Yes 4 Write out trees onto tree file? Yes Y to accept these or type the letter for one to change

The U, J, O, M, and 0 options are the usual ones, described in the main documentation file. The user trees for option U are read from a file whose default name is intree. The I option selects between Interleaved and Sequential input data formats, and is described in the documentation file for the molecular sequences programs.

The G (global search) option causes, after the last species is added to the tree, each possible group to be removed and re-added. This improves the result, since the position of every species is reconsidered. It approximately triples the run-time of the program.

The two options specific to this program are the A, and L options. The L (Length) option allows the user to specify the length in bases of the restriction sites. Allowed values are 1 to 8 (the constant "maxcutter" in file phylip.h controls the maximum allowed value). At the moment the program assumes that all sites have the same length (for example, that all enzymes are 6-base-cutters). The default value for this parameter is 6, which will be used if the L option is not invoked. A desirable future development for the package would be allowing the L parameter to be different for every site. It would also be desirable to allow for ambiguities in the recognition site, since some enzymes recognize 2 or 4 sequences. Both of these would require fairly complicated programming or else slower execution times.

The A (All) option specifies that all sites are detected, even those for which all of the species have the recognition sequence absent (character state "-"). The default condition is that it is assumed that such sites will not occur in the data. The likelihood computed when the A option is not used is the probability of the pattern of sites given that tree and conditional on the pattern not being all absences. This will be realistic for most data, except for cases in which the data are extracted from sites data for a larger number of species, in which case some of the site positions could have all absences in the subset of species. In such cases an effective way of analyzing the data would be to omit those sites and not use the A option, as such positions, even if not absolutely excluded, are nevertheless less likely than random to have been incorporated in the data set.

The W (Weights) option, which is invoked in the input file rather than in the menu, allows the user to select a subset of sites to be analyzed. It is invoked in the usual way, except that only weights 0 and 1 are allowed. If the W option is not used, all sites will be analyzed. If the Weights option is used, there must be a W in the first line of the input file.

OUTPUT FORMAT

The output starts by giving the number of species, and the number of sites. If the default condition is used instead of the A option the program states that it is assuming that sites absent in all species have been omitted. The value of the site length (6 bases, for example) is also given.

If option 2 (print out data) has been selected, there then follow the restriction site sequences, printed in groups of ten sites. The trees found are printed as an unrooted tree topology (possibly rooted by outgroup if so requested). The internal nodes are numbered arbitrarily for the sake of identification. The number of trees evaluated so far and the log likelihood of the tree are also given.

A table is printed showing the length of each tree segment, as well as (very) rough confidence limits on the length. As with DNAML, if a confidence limit is negative, this indicates that rearrangement of the tree in that region is not excluded, while if both limits are positive, rearrangement is still not necessarily excluded because the variance calculation on which the confidence limits are based results in an underestimate, which makes the confidence limits too narrow.

In addition to the confidence limits, the program performs a crude Likelihood Ratio Test (LRT) for each branch of the tree. The program computes the ratio of likelihoods with and without this branch length forced to zero length. This done by comparing the likelihoods changing only that branch length. A truly correct LRT would force that branch length to zero and also allow the other branch lengths to adjust to that. The result would be a likelihood ratio closer to 1. Therefore the present LRT will err on the side of being too significant.

One should also realize that if you are looking not at a previously-chosen branch but at all branches, that you are seeing the results of multiple tests. With 20 tests, one is expected to reach significance at the P = .05 level purely by chance. You should therefore use a much more conservative significance level, such as .05 divided by the number of tests. The significance of these tests is shown by printing asterisks next to the confidence interval on each branch length. It is important to keep in mind that both the confidence limits and the tests are very rough and approximate, and probably indicate more significance than they should. Nevertheless, maximum likelihood is one of the few methods that can give you any indication of its own error; most other methods simply fail to warn the user that there is any error! (In fact, whole philosophical schools of taxonomists exist whose main point seems to be that there isn't any error, that the "most parsimonious" tree is the best tree by definition and that's that).

The log likelihood printed out with the final tree can be used to perform various likelihood ratio tests. Remember that testing one tree topology against another is not a simple matter, because two different tree topologies are not hypotheses that are nested one within the other. If the trees differ by only one branch swap, it seems to be conservative to test the difference between their likelihoods with one degree of freedom, but other than that little is known and more work on this is needed.

If the U (User Tree) option is used and more than one tree is supplied, and the program is not told to assume autocorrelation between the rates at different sites, the program also performs a statistical test of each of these trees against the one with highest likelihood. If there are two user trees, the test done is one which is due to Kishino and Hasegawa (1989), a version of a test originally introduced by Templeton (1983). In this implementation it uses the mean and variance of log-likelihood differences between trees, taken across sites. If the two trees' means are more than 1.96 standard deviations different then the trees are declared significantly different. This use of the empirical variance of log-likelihood differences is more robust and nonparametric than the classical likelihood ratio test, and may to some extent compensate for the any lack of realism in the model underlying this program.

If there are more than two trees, the test done is an extension of the KHT test, due to Shimodaira and Hasegawa (1999). They pointed out that a correction for the number of trees was necessary, and they introduced a resampling method to make this correction. In the version used here the variances and covariances of the sum of log likelihoods across sites are computed for all pairs of trees. To test whether the difference between each tree and the best one is larger than could have been expected if they all had the same expected log-likelihood, log-likelihoods for all trees are sampled with these covariances and equal means (Shimodaira and Hasegawa's "least favorable hypothesis"), and a P value is computed from the fraction of times the difference between the tree's value and the highest log-likelihood exceeds that actually observed. Note that this sampling needs random numbers, and so the program will prompt the user for a random number seed if one has not already been supplied. With the two-tree KHT test no random numbers are used.

In either the KHT or the SH test the program prints out a table of the log-likelihoods of each tree, the differences of each from the highest one, the variance of that quantity as determined by the log-likelihood differences at individual sites, and a conclusion as to whether that tree is or is not significantly worse than the best one.

The branch lengths printed out are scaled in terms of expected numbers of base substitutions, not counting replacements of a base by itself. Of course, when a branch is twice as long this does not mean that there will be twice as much net change expected along it, since some of the changes occur in the same site and overlie or even reverse each other. Confidence limits on the branch lengths are also given. Of course a negative value of the branch length is meaningless, and a confidence limit overlapping zero simply means that the branch length is not necessarily significantly different from zero. Because of limitations of the numerical algorithm, branch length estimates of zero will often print out as small numbers such as 0.00001. If you see a branch length that small, it is really estimated to be of zero length.

Another possible source of confusion is the existence of negative values for the log likelihood. This is not really a problem; the log likelihood is not a probability but the logarithm of a probability, and since probabilities never exceed 1.0 this logarithm will typically be negative. The log likelihood is maximized by being made more positive: -30.23 is worse than -29.14. The log likelihood will not always be negative since a combinatorial constant has been left out of the expression for the likelihood. This does not affect the tree found or the likelihood ratios (or log likelihood differences) between trees.

THE ALGORITHM

The program uses a Newton-Raphson algorithm to update one branch length at a time. This is faster than the EM algorithm which was described in my paper on restriction sites maximum likelihood (Felsenstein, 1992). The likelihood that is being maximized is the same one used by Smouse and Li (1987) extended for multiple species. moving down on the likelihood surface. You may have to "tune" the value of extrapol to suit your data.

PROGRAM CONSTANTS

The constants include "maxcutter" (set in phylip.h), the maximum length of an enzyme recognition site. The memory used by the program will be approximately proportional to this value, which is 8 in the distribution copy. The program also uses constants "iterations" and "smoothings", and decreasing "epsilon". Reducing "iterations" and "smoothings" or increasing "epsilon" will result in faster execution but a worse result. These values will not usually have to be changed.

The program spends most of its time doing real arithmetic. The algorithm, with separate and independent computations occurring at each site, lends itself readily to parallel processing.

A feature of the algorithm is that it saves time by recognizing sites at which the pattern of presence/absence is the same, and does that computation only once. Thus if we have only four species but a large number of sites, there are only about (ignoring ambiguous bases) 16 different patterns of presence/absence (2 x 2 x 2 x 2) that can occur. The program automatically counts occurrences of each and does the computation for each pattern only once, so that it only needs to do as much computation as would be needed with at most 16 sites, even though the number of sites is actually much larger. Thus the program will run very effectively with few species and many sites.

PAST AND FUTURE OF THE PROGRAM

This program was developed by modifying DNAML version 3.1 and also adding some of the modifications that were added to DNAML version 3.2, with which it shares many of its data structures and much of its strategy. Version 3.6 changed from EM iterations of branch lengths, which involved arbitrary extrapolation factors, to the Newton-Raphson algorithm, which improved the speed of the program (though only from "very slow" to "slow").

There are a number of obvious directions in which the program needs to be modified in the future. Extension to allow for different rates of transition and transversion is straightforward, but would slow down the program considerably, as I have mentioned above. I have not included in the program any provision for saving and printing out multiple trees tied for highest likelihood, in part because an exact tie is unlikely.

TEST DATA SET

5 13 2 Alpha ++-+-++--+++- Beta ++++--+--+++- Gamma -+--+-++-+-++ Delta ++-+----++--- Epsilon ++++----++---

CONTENTS OF OUTPUT FILE (if all numerical options are on)

Restriction site Maximum Likelihood method, version 3.6 5 Species, 13 Sites, 2 Enzymes Recognition sequences all 6 bases long Sites absent from all species are assumed to have been omitted Name Sites ---- ----- Alpha ++-+-++--+ ++- Beta ++++--+--+ ++- Gamma -+--+-++-+ -++ Delta ++-+----++ --- Epsilon ++++----++ --- +----Gamma | | +Epsilon | +--3 1--2 +Delta | | | +Beta | +Alpha remember: this is an unrooted tree! Ln Likelihood = -40.34358 Between And Length Approx. Confidence Limits ------- --- ------ ------- ---------- ------ 1 Gamma 0.10813 ( 0.01154, 0.21901) ** 1 2 0.01156 ( zero, 0.04578) 2 3 0.05885 ( zero, 0.12697) ** 3 Epsilon 0.00100 ( zero, 0.00617) 3 Delta 0.01460 ( zero, 0.05036) 2 Beta 0.00100 ( zero, infinity) 1 Alpha 0.01310 ( zero, 0.04806) * = significantly positive, P < 0.05 ** = significantly positive, P < 0.01