fdnainvar

Wiki

Please help by correcting and extending the Wiki pages.

Function

Description

Algorithm

This program reads in nucleotide sequences for four species and computes the phylogenetic invariants discovered by James Cavender (Cavender and Felsenstein, 1987) and James Lake (1987). Lake's method is also called by him "evolutionary parsimony". I prefer Cavender's more mathematically precise term "invariants", as the method bears somewhat more relationship to likelihood methods than to parsimony. The invariants are mathematical formulas (in the present case linear or quadratic) in the EXPECTED frequencies of site patterns which are zero for all trees of a given tree topology, irrespective of branch lengths. We can consider at a given site that if there are no ambiguities, we could have for four species the nucleotide patterns (considering the same site across all four species) AAAA, AAAC, AAAG, ... through TTTT, 256 patterns in all.

The invariants are formulas in the expected pattern frequencies, not the observed pattern frequencies. When they are computed using the observed pattern frequencies, we will usually find that they are not precisely zero even when the model is correct and we have the correct tree topology. Only as the number of nucleotides scored becomes infinite will the observed pattern frequencies approach their expectations; otherwise, we must do a statistical test of the invariants.

Some explanation of invariants will be found in the above papers, and also in my recent review article on statistical aspects of inferring phylogenies (Felsenstein, 1988b). Although invariants have some important advantages, their validity also depends on symmetry assumptions that may not be satisfied. In the discussion below suppose that the possible unrooted phylogenies are I: ((A,B),(C,D)), II: ((A,C),(B,D)), and III: ((A,D),(B,C)).

Lake's Invariants, Their Testing and Assumptions

Lake's invariants are fairly simple to describe: the patterns involved are only those in which there are two purines and two pyrimidines at a site. Thus a site with AACT would affect the invariants, but a site with AAGG would not. Let us use (as Lake does) the symbols 1, 2, 3, and 4, with the proviso that 1 and 2 are either both of the purines or both of the pyrimidines; 3 and 4 are the other two nucleotides. Thus 1 and 2 always differ by a transition; so do 3 and 4. Lake's invariants, expressed in terms of expected frequencies, are the three quantities:

(1)      P(1133) + P(1234) - P(1134) - P(1233), 

(2)      P(1313) + P(1324) - P(1314) - P(1323), 

(3)      P(1331) + P(1342) - P(1341) - P(1332), 

He showed that invariants (2) and (3) are zero under Topology I, (1) and (3) are zero under topology II, and (1) and (2) are zero under Topology III. If, for example, we see a site with pattern ACGC, we can start by setting 1=A. Then 2 must be G. We can then set 3=C (so that 4 is T). Thus its pattern type, making those substitutions, is 1323. P(1323) is the expected probability of the type of pattern which includes ACGC, TGAG, GTAT, etc.

Lake's invariants are easily tested with observed frequencies. For example, the first of them is a test of whether there are as many sites of types 1133 and 1234 as there are of types 1134 and 1233; this is easily tested with a chi-square test or, as in this program, with an exact binomial test. Note that with several invariants to test, we risk overestimating the significance of results if we simply accept the nominal 95% levels of significance (Li and Guoy, 1990).

Lake's invariants assume that each site is evolving independently, and that starting from any base a transversion is equally likely to end up at each of the two possible bases (thus, an A undergoing a transversion is equally likely to end up as a C or a T, and similarly for the other four bases from which one could start. Interestingly, Lake's results do not assume that rates of evolution are the same at all sites. The result that the total of 1133 and 1234 is expected to be the same as the total of 1134 and 1233 is unaffected by the fact that we may have aggregated the counts over classes of sites evolving at different rates.

Cavender's Invariants, Their Testing and Assumptions

Cavender's invariants (Cavender and Felsenstein, 1987) are for the case of a character with two states. In the nucleic acid case we can classify nucleotides into two states, R and Y (Purine and Pyrimidine) and then use the two-state results. Cavender starts, as before, with the pattern frequencies. Coding purines as R and pyrimidines as Y, the patterns types are RRRR, RRRY, and so on until YYYY, a total of 16 types. Cavender found quadratic functions of the expected frequencies of these 16 types that were expected to be zero under a given phylogeny, irrespective of branch lengths. Two invariants (called K and L) were found for each tree topology. The L invariants are particularly easy to understand. If we have the tree topology ((A,B),(C,D)), then in the case of two symmetric states, the event that A and B have the same state should be independent of whether C and D have the same state, as the events determining these happen in different parts of the tree. We can set up a contingency table:

                                 C = D         C =/= D
                           ------------------------------
                          |
                   A = B  |   YYYY, YYRR,     YYYR, YYRY,
                          |   RRRR, RRYY      RRYR, RRRY
                          |
                 A =/= B  |   YRYY, YRRR,     YRYR, YRRY,
                          |   RYYY, RYRR      RYYR, RYRY

and we expect that the events C = D and A = B will be independent. Cavender's L invariant for this tree topology is simply the negative of the crossproduct difference,

      P(A=/=B and C=D) P(A=B and C=/=D) - P(A=B and C=D) P(A=/=B and C=/=D). 

One of these L invariants is defined for each of the three tree topologies. They can obviously be tested simply by doing a chi-square test on the contingency table. The one corresponding to the correct topology should be statistically indistinguishable from zero. Again, there is a possible multiple tests problem if all three are tested at a nominal value of 95%.

The K invariants are differences between the L invariants. When one of the tables is expected to have crossproduct difference zero, the other two are expected to be nonzero, and also to be equal. So the difference of their crossproduct differences can be taken; this is the K invariant. It is not so easily tested.

The assumptions of Cavender's invariants are different from those of Lake's. One obviously need not assume anything about the frequencies of, or transitions among, the two different purines or the two different pyrimidines. However one does need to assume independent events at each site, and one needs to assume that the Y and R states are symmetric, that the probability per unit time that a Y changes into an R is the same as the probability that an R changes into a Y, so that we expect equal frequencies of the two states. There is also an assumption that all sites are changing between these two states at the same expected rate. This assumption is not needed for Lake's invariants, since expectations of sums are equal to sums of expectations, but for Cavender's it is, since products of expectations are not equal to expectations of products.

It is helpful to have both sorts of invariants available; with further work we may appreciate what other invaraints there are for various models of nucleic acid change.

Usage

% fdnainvar -printdata Nucleic acid sequence Invariants method Input (aligned) nucleotide sequence set(s): dnainvar.dat Phylip weights file (optional): Phylip dnainvar program output file [dnainvar.fdnainvar]: Output written to output file "dnainvar.fdnainvar" Done.

Go to the input files for this example
Go to the output files for this example

Command line arguments

Nucleic acid sequence Invariants method
Version: EMBOSS:6.4.0.0

   Standard (Mandatory) qualifiers:
  [-sequence]          seqsetall  File containing one or more sequence
                                  alignments
   -weights            properties Phylip weights file (optional)
  [-outfile]           outfile    [*.fdnainvar] Phylip dnainvar program output
                                  file

   Additional (Optional) qualifiers (* if not always prompted):
   -printdata          boolean    [N] Print data at start of run
*  -[no]dotdiff        boolean    [Y] Use dot-differencing to display results
   -[no]printpattern   boolean    [Y] Print counts of patterns
   -[no]printinvariant boolean    [Y] Print invariants
   -[no]progress       boolean    [Y] Print indications of progress of run

   Advanced (Unprompted) qualifiers: (none)
   Associated qualifiers:

   "-sequence" associated qualifiers
   -sbegin1            integer    Start of each sequence to be used
   -send1              integer    End of each sequence to be used
   -sreverse1          boolean    Reverse (if DNA)
   -sask1              boolean    Ask for begin/end/reverse
   -snucleotide1       boolean    Sequence is nucleotide
   -sprotein1          boolean    Sequence is protein
   -slower1            boolean    Make lower case
   -supper1            boolean    Make upper case
   -sformat1           string     Input sequence format
   -sdbname1           string     Database name
   -sid1               string     Entryname
   -ufo1               string     UFO features
   -fformat1           string     Features format
   -fopenfile1         string     Features file name

   "-outfile" associated qualifiers
   -odirectory2        string     Output directory

   General qualifiers:
   -auto               boolean    Turn off prompts
   -stdout             boolean    Write first file to standard output
   -filter             boolean    Read first file from standard input, write
                                  first file to standard output
   -options            boolean    Prompt for standard and additional values
   -debug              boolean    Write debug output to program.dbg
   -verbose            boolean    Report some/full command line options
   -help               boolean    Report command line options and exit. More
                                  information on associated and general
                                  qualifiers can be found with -help -verbose
   -warning            boolean    Report warnings
   -error              boolean    Report errors
   -fatal              boolean    Report fatal errors
   -die                boolean    Report dying program messages
   -version            boolean    Report version number and exit

Qualifier	Type	Description	Allowed values	Default
Standard (Mandatory) qualifiers
[-sequence] (Parameter 1)	seqsetall	File containing one or more sequence alignments	Readable sets of sequences	Required
-weights	properties	Phylip weights file (optional)	Property value(s)
[-outfile] (Parameter 2)	outfile	Phylip dnainvar program output file	Output file	<*>.fdnainvar
Additional (Optional) qualifiers
-printdata	boolean	Print data at start of run	Boolean value Yes/No	No
-[no]dotdiff	boolean	Use dot-differencing to display results	Boolean value Yes/No	Yes
-[no]printpattern	boolean	Print counts of patterns	Boolean value Yes/No	Yes
-[no]printinvariant	boolean	Print invariants	Boolean value Yes/No	Yes
-[no]progress	boolean	Print indications of progress of run	Boolean value Yes/No	Yes
Advanced (Unprompted) qualifiers
(none)
Associated qualifiers
"-sequence" associated seqsetall qualifiers
-sbegin1 -sbegin_sequence	integer	Start of each sequence to be used	Any integer value	0
-send1 -send_sequence	integer	End of each sequence to be used	Any integer value	0
-sreverse1 -sreverse_sequence	boolean	Reverse (if DNA)	Boolean value Yes/No	N
-sask1 -sask_sequence	boolean	Ask for begin/end/reverse	Boolean value Yes/No	N
-snucleotide1 -snucleotide_sequence	boolean	Sequence is nucleotide	Boolean value Yes/No	N
-sprotein1 -sprotein_sequence	boolean	Sequence is protein	Boolean value Yes/No	N
-slower1 -slower_sequence	boolean	Make lower case	Boolean value Yes/No	N
-supper1 -supper_sequence	boolean	Make upper case	Boolean value Yes/No	N
-sformat1 -sformat_sequence	string	Input sequence format	Any string
-sdbname1 -sdbname_sequence	string	Database name	Any string
-sid1 -sid_sequence	string	Entryname	Any string
-ufo1 -ufo_sequence	string	UFO features	Any string
-fformat1 -fformat_sequence	string	Features format	Any string
-fopenfile1 -fopenfile_sequence	string	Features file name	Any string
"-outfile" associated outfile qualifiers
-odirectory2 -odirectory_outfile	string	Output directory	Any string
General qualifiers
-auto	boolean	Turn off prompts	Boolean value Yes/No	N
-stdout	boolean	Write first file to standard output	Boolean value Yes/No	N
-filter	boolean	Read first file from standard input, write first file to standard output	Boolean value Yes/No	N
-options	boolean	Prompt for standard and additional values	Boolean value Yes/No	N
-debug	boolean	Write debug output to program.dbg	Boolean value Yes/No	N
-verbose	boolean	Report some/full command line options	Boolean value Yes/No	Y
-help	boolean	Report command line options and exit. More information on associated and general qualifiers can be found with -help -verbose	Boolean value Yes/No	N
-warning	boolean	Report warnings	Boolean value Yes/No	Y
-error	boolean	Report errors	Boolean value Yes/No	Y
-fatal	boolean	Report fatal errors	Boolean value Yes/No	Y
-die	boolean	Report dying program messages	Boolean value Yes/No	Y
-version	boolean	Report version number and exit	Boolean value Yes/No	N

Input file format

Input files for usage example

File: dnainvar.dat

   4   13
Alpha     AACGTGGCCAAAT
Beta      AAGGTCGCCAAAC
Gamma     CATTTCGTCACAA
Delta     GGTATTTCGGCCT

Output file format

If option 3 is on the invariants are then printed out, together with their statistical tests. For Lake's invariants the two sums which are expected to be equal are printed out, and then the result of an one-tailed exact binomial test which tests whether the difference is expected to be this positive or more. The P level is given (but remember the multiple-tests problem!).

For Cavender's L invariants the contingency tables are given. Each is tested with a one-tailed chi-square test. It is possible that the expected numbers in some categories could be too small for valid use of this test; the program does not check for this. It is also possible that the chi-square could be significant but in the wrong direction; this is not tested in the current version of the program. To check it beware of a chi-square greater than 3.841 but with a positive invariant. The invariants themselves are computed, as the difference of cross-products. Their absolute magnitudes are not important, but which one is closest to zero may be indicative. Significantly nonzero invariants should be negative if the model is valid. The K invariants, which are simply differences among the L invariants, are also printed out without any test on them being conducted. Note that it is possible to use the bootstrap utility SEQBOOT to create multiple data sets, and from the output from sunning all of these get the empirical variability of these quadratic invariants.

Output files for usage example

File: dnainvar.fdnainvar

Nucleic acid sequence Invariants method, version 3.69

 4 species,  13  sites

Name            Sequences
----            ---------

Alpha        AACGTGGCCA AAT
Beta         ..G..C.... ..C
Gamma        C.TT.C.T.. C.A
Delta        GGTA.TT.GG CC.



   Pattern   Number of times

     AAAC         1
     AAAG         2
     AACC         1
     AACG         1
     CCCG         1
     CCTC         1
     CGTT         1
     GCCT         1
     GGGT         1
     GGTA         1
     TCAT         1
     TTTT         1


Symmetrized patterns (1, 2 = the two purines  and  3, 4 = the two pyrimidines
                  or  1, 2 = the two pyrimidines  and  3, 4 = the two purines)

     1111         1
     1112         2
     1113         3
     1121         1
     1132         2
     1133         1
     1231         1
     1322         1
     1334         1

Tree topologies (unrooted): 

    I:  ((Alpha,Beta),(Gamma,Delta))
   II:  ((Alpha,Gamma),(Beta,Delta))
  III:  ((Alpha,Delta),(Beta,Gamma))



  [Part of this file has been deleted for brevity]

different purine:pyrimidine ratios from 1:1.

  Tree I:

   Contingency Table

      2     8
      1     2

   Quadratic invariant =             4.0

   Chi-square =    0.23111 (not significant)


  Tree II:

   Contingency Table

      1     5
      1     6

   Quadratic invariant =            -1.0

   Chi-square =    0.01407 (not significant)


  Tree III:

   Contingency Table

      1     2
      6     4

   Quadratic invariant =             8.0

   Chi-square =    0.66032 (not significant)




Cavender's quadratic invariants (type K) using purines vs. pyrimidines
 (these are expected to be zero for the correct tree topology)
They will be misled if there are substantially
different evolutionary rate between sites, or
different purine:pyrimidine ratios from 1:1.
No statistical test is done on them here.

  Tree I:              -9.0
  Tree II:              4.0
  Tree III:             5.0

Data files

Notes

References

Warnings

Diagnostic Error Messages

Exit status

Known bugs

See also

Author(s)

Please report all bugs to the EMBOSS bug team (emboss-bug © emboss.open-bio.org) not to the original author.

History

Converted (August 2004) to an EMBASSY program by the EMBOSS team.

Target users