Copyright 1990, 1994, 1996 Gerald Z. Hertz
May be copied for noncommercial purposes.
Author:
Gerald Z. Hertz
Dept. of Molecular, Cellular, and Developmental Biology
University of Colorado
Campus Box 347
Boulder, CO 80309-0347
hertz@boulder.colorado.edu
CONSENSUS (version 4b)
This program determines consensus patterns in unaligned sequences. The
algorithm is based on a matrix representation of a consensus pattern. Each
row corresponds to one of the letters of the relevant alphabet---e.g., 4
rows in the case of DNA. Each column corresponds to one of the positions
within the pattern. The elements of the matrix are determined by the
number of times that the indicated letter occurs at the indicated position.
Matrices are constructed by sequentially adding additional L-mers
(subsequences of length L, where L is the width of the pattern being
sought) to previously saved matrices. During each cycle, only the
most significant matrices are saved. The maximum number of matrices to
save is determined by the "-q" option (see section 1 below). In
practice, less matrices are ultimately saved because many of the
matrices initially saved are identical to each other.
The program can use 3 different criteria for deciding to stop adding
additional words to the saved matrices:
1) Each sequence has contributed exactly one word to the saved
matrices (the default).
2) The saved matrices contain a maximum allowable number of words (set
with the -n and -N options).
3) The program has completed a designated number of cycles since finding
the current most significant alignment (set with the -t option).
This latter criteria is used in addition to criteria 1) and 2)
to terminate the program sooner.
The significance of a matrix is initially measured by its information
content. A higher information content indicates a rarer pattern and a
more desirable matrix. The information content of a matrix pattern
can be used to predict how frequently a pattern is expected to occur
in a random sequence, and to estimate its statistical significance
(i.e. p-value). A matrix's statistical significance is obtained from
the log-likelihood interpretation of the information content.
Currently, we approximate statistical significance using a stretched
out chi-square distribution so that the average information matches
the average of the chi-square distribution and the maximal possible
information has the correct probability. When comparing matrices
summarizing differing numbers of sequences and having differing
widths, it is also necessary to multiply the statistical significance
by the approximate number of possible alignments, containing the
designated number of sequences and having the observed width. We
refer to this product as the expected frequency of the matrix
alignment.
UNFORTUNATELY, OUR P-VALUE CALCULATIONS---AND, THUS, THE EXPECTED
FREQUENCY---ARE VERY APPROXIMATE AND ULTIMATELY NEED TO BE DETERMINED
BY ALIGNING RANDOMIZED SEQUENCES. The relative p-values of different
alignments are presumably more reliable than the absolute p-values.
The program can print two different lists of matrices. The first list
contains the matrices having the highest information content from each
cycle, ordered by decreasing statistical significance (i.e.,
increasing expected frequency). In general, this first list will
contain the most interesting alignment. The second list contains the
matrices saved after the final cycle of the program, also ordered by
decreasing statistical significance. In general, this latter list
will be useful when the user wishes each sequence to contribute
exactly one word to the final alignment (i.e., when the -n and -N
options are not used).
In the program's output, the words contained in each matrix are listed
in the order of their occurrence in the input sequences. The order is
indicated by "integer|integer". The first integer is simply a
sequential count of the words, and the second integer indicates during
which cycle the word was added to the matrix. The location of a word
is indicated by "integer/integer". The first integer indicates which
sequence contains the word, and the second integer indicates where in
that sequence the word is located. If the first integer is preceded
by a minus sign, then the complementary word is the one included in
the matrix.
The output of the program is sent to the standard output. The input
files---those containing the actual sequences and those indicated by
the "-f", "-a", and "-i" options---can contain comments according to
the following convention. The portion of a line following a ';', '%',
or '#' is considered a comment and is ignored. Comments can begin
anywhere in a line and always end at the end of the line. The one
minor exception is that, to avoid ambiguity, comments in the list of
sequences (see the "-f" option below) must be preceded by a blank
space when not occurring at the beginning of a line.
COMMAND LINE OPTIONS:
0) -h: print these directions.
1) General information
-f filename: this file (default: read from the standard input) contains
the names of the sequences. The names of the sequences must be
less than 512 characters. The corresponding sequence may follow
its name if the sequence is enclosed between backslashes (\).
Otherwise, the sequence is assumed to be in a separate file having
the indicated name. The format of the actual sequences is described
at the end of these directions. The following four modifiers can
appear in front of the name of the relevant sequence:
-c: the sequence is circular.
-s integer-integer integer-integer: the positions in the sequence
indicated by the integer pairs, inclusive, are seed sequences.
If the "-s" modifier is used anywhere in the input file, then the
initial set of matrices will only be constructed (i.e., seeded)
from the sequences within the marked regions. If this modifier
is not used anywhere in the input file, then all the sequences
will be used to seed matrices. One or more integer pair can be
indicated for a single sequence. However, if no integer pairs
are given, the whole sequence will be used for seeding matrices.
-i integer-integer integer-integer: the positions in the sequence
indicated by the integer pairs, inclusive, are the only positions
to be analyzed.
-e integer-integer integer-integer: the positions in the sequence
indicated by the integer pairs, inclusive, are to be excluded
from the analysis.
When both the "-i" and "-e" modifiers are used, the intersection
of permissible positions is analyzed. When a sequence name is
not marked by either the "-i" or "-e" modifier, then the whole
sequence is included in the analysis.
-L integer: width of the pattern being sought (default: 20).
-q integer: the maximum number of matrices to save between cycles of the
program---i.e., the queue size (default: save 1000 matrices).
2) Alphabet options
-d: use the designated prior probabilities of the letters to override the
observed frequencies. By default, the program uses the frequencies
observed in your own sequence data for the prior probabilities of the
letters. However, if the "-d" option is set, the prior probabilities
designated by one of the next 3 options are used. If the "-d" option
is not set, the next 3 options are still used to determine the
sequence alphabet, but any prior probability information is ignored.
The next three options are mutually exclusive (default: "-a alphabet").
-a filename: file containing the alphabet and normalization information.
[Use "-af" when using the VMS operating system]
Each line contains a letter (a symbol in the alphabet) followed by an
optional normalization number (default: 1.0). The normalization is
based on the relative prior probabilities of the letters. For nucleic
acids, this might be be the genomic frequency of the bases; however,
if the "-d" option is not used, the frequencies observed in your own
sequence data are used. In nucleic acid alphabets, a letter and its
complement appear on the same line, separated by a colon (a letter can
be its own complement, e.g. when using a dimer alphabet).
Complementary letters may use the same normalization number. Only the
standard 26 letters are permissible; however, when the "-CS" option is
used, the alphabet is case sensitive so that a total of 52 different
characters are possible.
POSSIBLE LINE FORMATS WITHOUT COMPLEMENTARY LETTERS:
letter
letter normalization
POSSIBLE LINE FORMATS WITH COMPLEMENTARY LETTERS:
letter:complement
letter:complement normalization
letter:complement normalization:complement's_normalization
-i filename: same as the "-a" option, except that the symbols of
the alphabet are represented by integers rather than by letters.
Any integer permitted by the machine is a permissible symbol.
[Use "-if" when using the VMS operating system]
-A alphabet_and_normalization_information: same as "-a" option, except
information appears on the command line (e.g., -A a:t 3 c:g 2).
[Use "-ac" when using the VMS operating system]
3) Alphabet modifier indicating whether ascii alphabets are case
sensitive---the following option is mutually exclusive with
the "-i" option (default: ascii alphabets are case insensitive).
-CS: ascii alphabets are case sensitive.
[Use "-as" when using the VMS operating system]
4) Options for handling the complement of nucleic acid sequences---
the four options in this section are mutually exclusive
-c0: ignore the complement (the default option)
-c1: include both strands as separate sequences
-c2: include both strands as a single sequence (i.e., orientation unknown)
-c3: assume pattern is symmetrical
5) Algorithm options
the "-l", "-n", and "-N" options are mutually exclusive;
the "-m" option can only be used when the "-n" or "-N" option is used.
-l: (lowercase L) seed with the first sequence and proceed linearly
through the list. This option results in a significant speed
up in the program, but the algorithm becomes dependent on the
order of the sequence-file names. This option corresponds to
the original "consensus" algorithm (Stormo and Hartzell, 1989,
PNAS, 86:1183-1187; Hertz et al., 1990, CABIOS, 6:81-92).
-n integer: repeat the matrix building cycle a maximum of "integer"
times and allow each sequence to contribute zero or more words
per matrix. [Use "-n1" when using the VMS operating system]
-N integer: repeat the matrix building cycle a maximum of "integer"
times and allow each sequence to contribute one or more words
per matrix. [Use "-n2" when using the VMS operating system]
-m integer: the minimum distance between the starting points of words
within the same matrix pattern; must be a positive integer; can only
be used when the "-n" or "-N" option is also used. If the integer
is a 1, then there is no restriction on the overlap. If the integer
is the same as the integer indicated by the "-L" option, then no
overlap is allowed (default: integer indicated by the "-L" option).
When the "-c2" option is used (see below), then the "-m" option also
indicates the minimum distance between the start of a word and the
end of a word on the complementary strand.
-t integer: terminate the program "integer" cycles after the current
most significant alignment is identified (default: terminate only
when the maximum number of matrix building cycles is completed).
The "-q", "-n", and "-t" options can be changed after the program
starts by placing the new options in a file called "changes." suffixed
with the process identification number---the PID number listed at the
beginning of the program's output. For example a file called
"changes.10568" might contain "-q10 -n50 -t2". The "-n" option
can change the maximum number of words in the alignments even if it
was not used at the beginning of the program, although it will not
permit a sequence to contribute more than one word to the alignment
unless the "-n" or "-N" option was used on the command line. If
the "-t" option was not used when the program was started, this option
will only keep track of alignments beginning with the cycle during
which it is first initiated.
6) Output options
-pt integer: the number of matrices to print of the top matrices from each
cycle (default: 4). A negative value means print all the top matrices.
-pf integer: the number of matrices to print of the matrices saved from
the final cycle (default when NOT using "-n" option: print 4 matrices;
default when using "-n" option: print no matrices).
FORMAT OF THE SEQUENCE FILES
Do not explicitly give the complements of nucleic acid sequences. If
needed, the complementary sequence is determined by the program.
Whitespace, periods, dashes (unless part of an integer when the "-i"
option is used), and comments beginning with ';', '%', or '#' are
ignored. When using letter characters (i.e., with the "-a" and "-A"
alphabet options), integers are also ignored so that the sequence file
can contain positional information. When using integer characters
(i.e., with the "-i" alphabet option) the integers must be separated
by whitespace.
Sequences surrounded by slashes (/) do not contribute to the
generation of the patterns; thus, a portion of a sequence can be
ignored without disrupting the overall numbering of the sequence.
A double slash (//) would indicate a discontinuity in the sequence.
A '/' at the beginning or the end of a sequence will cause the sequence
to be marked as non-circular even if the sequence's name is marked
with a "-c" (see the "-f" option in section 1). The effect of the
single slashes can also be created with the "-i" and "-e" modifiers in
the file containing the names of the sequences (see the "-f" option in
section 1). When slashes and the "-i" and "-e" modifiers are all
used, the intersection of permissible positions is analyzed.
Sequences that follow their name in the file indicated by the "-f"
option must be enclosed between backslashes (\) (i.e., the actual
sequence must be preceded and followed by a backslash). However, if
the sequence is contained in a separate file, do NOT use a '\'.