Title
sqom -- Optimal matching of sequences
Syntax
sqom [if] [in] [, options]
options Description ------------------------------------------------------------------------- indelcost(#) set indel costs to # subcost(#|implied formula|matexp|matname) specify substitution costs name(varname) specify substitution costs refseqid(spec) select reference sequence full calculate full dissimilarity matrix between sequences k(#) restrict indels (to save calculation time) standard(#|cut|longer|longest|none) standardization of sequences of different length subsequence(a,b) use only subsequence between positions a and b idealtype(spec) compare with a specified ideal typical sequence -------------------------------------------------------------------------
Description
sqom performs optimal matching of sequences. The command uses the Needleman-Wunsch algorithm to find the alignment between two sequences that have the lowest Levenshtein distance. The Levenshtein distances are then stored for further use.
By default, all sequences are compared to the most frequent sequence and the resulting distances are stored in a variable. It is, however, possible to compare all sequences with a preselected reference distance or to compare all sequences with every other sequence. In the latter case, the resulting distances are stored in a Stata variable.
Comparing all sequences with any other sequence is computationally intensive.
Options
indelcost(#) specifies the cost attached to an insertion or deletion of an alignment. The default is indelcost(1).
subcost(#|implied formula|matexp|matname} specifies the cost attached to a substitution in an alignment. Substitution costs may be specified as real number, as implied formula, or as full matrix. Specifying substitution cost as, for example, subcost(3) will attach the cost of 3 to any substitution necessary in an alignment, regardless of how similar the substituted values may be. The default is two times the value specified as indel cost. A full substitution cost matrix can be specified either by specifying the name of a matrix containing the substitution cost or by typing valid matrix syntax into the option itself. The matrix has to be a symmetric n*n matrix, where n is the number of different elements in all sequences.
implied formula generates substitution costs based on the data. The implied formula is specified with a keyword. The following keywords are allowed
implied formula ------------------------------------------------------------------------- rawdistance use the absolute value of the difference between the numeric values representing the respective elements meanprobdistance calculates symmetric substitiution cost matrix based on the mean of the transitions' probabilities (p) in the data between every two neighboring elements in the sequences. The substitution costs between elements x and y are defined by: SC(x,y) = SC(y,x) = 2-p(x,y)-p(y,x) if x is not equal to y, otherwise 0. minprobdistance calculates symmetric substitiution cost matrix based on the transitions' probabilities (p) in the data between every two neighboring elements in sequences. The substitution cost matrix contains the minimal substitution costs for each pair of symmetric transitions: SC(x,y) = SC(y,x) = min(1-p(x,y),1-p(y,x))*2 if x is not equal to y, otherwise 0. maxprobdistance calculates symmetric substitiution cost matrix based on the transitions' probabilities (p) in the data between every two neighboring elements in sequences. The substitution cost matrix contains the maximal substitution costs for each pair of symmetric transitions: SC(x,y) = SC(y,x) = max(1-p(x,y),1-p(y,x))*2 if x is not equal to y, otherwise 0. ------------------------------------------------------------------------- The substitution costs in last three cases have values between 0 and 2
Specifying a full substitution cost matrix or generating a data based substitution cost matrix can increase the running time of the program considerably. Option k() might be considered for sqom with full substitution cost matrix.
name(varname) is used to specify the name of the variable in which the distances are stored. If not specified, _SQdist is used. The automatically generated distance variable will get overwritten without warning whenever a sqom command without name() is invoked.
refseqid(spec) is used to select the reference sequence against which all sequences in the dataset are being tested. Within the parentheses, an existing value of the sequence identifier has to be stated.
full is used to perform optimal matching for all sequences in the dataset against any other. The results of these comparisons are stored in the distance matrix "SQdist". Specifying the option full will increase the running time of the program considerably. Option k() might be used for sqom with full.
Two companion programs, sqclusterdat and sqclustermat, help to further analyze the distance matrix produced with sqom, full.
k(#) is used to speed up the calculation of the optimal matching algorithm. Within the parentheses, an integer positive number between 1 and the number of positions of the longest sequence can be given. The speed up will be higher with small numbers. Very small numbers can have the effect that the algorithm doesn't find the best alignment between some sequences, and this problem tends to increase if substitution costs are high relative to indel costs.
Note: The implementation of the k() is based partly on the source code of TDA, written by Goetz Rohwer and Ulrich Poetter. TDA is a very powerful program for transitory data analysis. It is programmed in C and distributed as freeware under the terms of the General Public License. It is downloadable from http://www.stat.ruhr-uni-bochum.de/tda.html.
standard(#|cut|longer|longest|none) is used to define the standardization of the resulting distances. With standard(#) all sequences are cut to the length #. The keyword cut automatically cuts all sequences to the length of the shortest sequence in the dataset. standard(longer) divides all distances by the length of the longer sequence of the respective alignment. standard(longest) divides all distances by the length of the longest sequence in the dataset; this is the default. none is specified if no standardization is needed.
subsequence(a,b) is used to include only the part of the sequence that is between position a and b, whereby a and b refer to the position defined in the order variable.
idealtype(spec) allows to specify an ideal typical sequence against which all sequences are compared. To specify the sequence use element[:repetitions] [element:repetitions]. For example, with idealtype(3:20 5 1:20 3:20) you specifiy an ideal typical sequence of length 61. The ideal typical sequence starts with element 3 over 20 positions, followed by one position of elment 5, 20 positions of element 1 and finally again 20 positions of element 3.
Authors
Ulrich Kohler, WZB, kohler@wz-berlin.de Magdalena Luniak, WZB, luniak@wz-berlin.de
Examples
. sqom, name(mydist) . sqindexplot, order(mydist)
. sqom, full k(2) . sqclustermat ward, name(mydist2) . sqindexplot, order(mydist2)
Also see
Online: sq, sqdemo, sqset, sqdes, sqegen, sqstat, sqindexplot, sqparcoord, sqom, sqclusterdat, sqclustermat