STAMP Help
Contents:

Citing STAMP:
Please cite one of the following papers. The second reference describes the algorithms and methods used in STAMP.

Input Formats:
STAMP allows motifs to be input in the following formats. Mixtures of different formats may be input. Note that the only acceptable formats are frequency or count matrices. Log-based probability weight matrices are not currently acceptable.
NameDescription and Recognition RulesExample
TRANSFAC STAMP recognizes TRANSFAC format matrices by the presence of "DE", "NA" or "P0" tags as the first words on a line. If the "DE" or "NA" tags are present, the motif name is taken as the next word on the line. The frequency matrix is read as the first 5 or 6-column line after the DE, NA or P0 line. All other TRANSFAC tags (AC, ID, BF, XX, etc.) before or after the matrix are ignored. The format of the matrix is:
Column1: Matrix position
Column2: A frequency
Column3: C frequency
Column4: G frequency
Column5: T frequency
Column6: Optional consensus letter
The recording of the frequency matrix stops at the next motif-format tag encountered or when the number of columns in a line drops below 5. 		NA Mync
XX
DE Mync
XX
P0 A C G T
01 0 31 0 0 C
02 29 0 0 2 A
03 0 30 0 1 C
04 2 1 28 0 G
05 0 3 0 28 T
06 0 0 31 0 G
XX
TRANSFAC-like This format is a simplification of the TRANSFAC format described above. The format of the matrix remains the same, but each matrix is directly preceded by a line beginning with a "DE" tag and followed by the motif name. This format is used by the SOMBRERO motif-finder. DE Mync
01 0 31 0 0 C
02 29 0 0 2 A
03 0 30 0 1 C
04 2 1 28 0 G
05 0 3 0 28 T
06 0 0 31 0 G
XX
Raw PSSM As with the TRANSFAC formats, this format represents motifs as Lx4 matrices. In the "Raw PSSM" format, the matrix is directly preceded by a line beginning with the > character. The name of the motif is taken as the next word after the > character. The matrix itself consists of a set of 4-column lines in the order A C G T. >Mync
0 31 0 0
29 0 0 2
0 30 0 1
2 1 28 0
0 3 0 28
0 0 31 0
Jaspar/Consite This format is used by the Jaspar database and Consite program. It is the only 4xL matrix format currently recognized by STAMP. The Jaspar motif must be preceded by a >-containing line which also contains the motif name. In addition, the four rows of the matrix must begin with the DNA letter represented by the frequencies in that row (i.e. A C G T). Square brackets "[]" may be present in the motif (as they are on the Jaspar webpages), but they will be ignored and are not required. > Mycn
A [ 0 29 0 2 0 0 ]
C [31 0 30 1 3 0 ]
G [ 0 0 0 28 0 31]
T [ 0 2 1 0 28 0 ]
MEME output The MEME motif format is recognized by the presence of a line beginning with "letter-probability". This line must be immediately followed by a 4-column frequency matrix in the order A C G T. ------------------------
Motif 2 position-specific probability matrix
------------------------
letter-probability matrix: alength= 4 w= 6 nsites= 31
0 31 0 0
29 0 0 2
0 30 0 1
2 1 28 0
0 3 0 28
0 0 31 0
Consensus sequence STAMP converts consensus sequences into probability matrices using literal IUPAC rules. The definition of degenerate characters are as follows:
M = A or C (i.e. A=0.5 C=0.5 G=0 T=0)
R = A or G
W = A or T
S = C or G
Y = C or T
K = G or T
V = not T (i.e. A=1/3 C=1/3 G=1/3 T=0)
H = not G
D = not C
B = not A
N = A=C=G=T=0.25
 >RUNX1
NNWCTYGGTY
Sequence Alignment STAMP allows the input of sequence alignments, from which a frequency matrix is constructed. Each input alignment must be preceded by a >-containing line that also has a name for the alignment. All sequences in the alignment must be the same length as each other. The aligned sequences may contain gap characters ("-" or ".") or consensus sequence letters. >Align1
AACACGTGGC
GCCACGT-CC
CGCATGTGCA
A--ACGTGTT
NACWCGTGCC
XMS XMS is an XML format for describing regulatory motifs and PSSMs. This format was defined by Thomas Down, and used in the NestedMICA and MotifExplorer programs. XMS sample

In addition, STAMP allows the input of entire output files from the following motif-finders. This selection is limited to 14 motif-finders at the moment, but may be easily extended in the future. Please email me (mahony(AT)mit.edu) with any suggestions for other motif-finder output files that may be suitable for input to STAMP. In the suggestion, please note any unique features at the beginning of the output file that will distinguish the motif-finder's format from others.
NameDescription and Recognition RulesExample
SOMBRERO The SOMBRERO output files are too large to be uploaded to STAMP. However, you can use this Perl script to convert the top X number of motifs found by SOMBRERO into an acceptable TRANSFAC-like format input file. Example of output from the TopX.pl script applied to the results of a SOMBRERO analysis.
BioProspector or CompareProspector Bioprospector output files are recognized if the first 5 lines of the file contain the string "BioProspector" and two lines later a string of "*******" appears. BioProspector motifs begin with a line starting with a "Blk1" or "Blk2" tag and the matrix consists of a set of 9-column lines, of which columns 2-5 are A C G T frequency counts. BioProspector output.
MDscan MDscan's motifs are returned in the same format as BioProspectors. However, MDscan uses two different file formats (one from the command line version and one in the emailed results from the online version). The command-line version output files are recognized via a series of unique keywords in the first 5 lines of the file ("Pm", "Mtf" and "Final Motif"). The email version is recognized via the tag "MDscan Search Result" followed two lines later by a string of "*******". MDscan output.
AlignACE AlignACE output files (from the command-line version of AlignACE) contain the string "AlignACE" as the first word in the file. The motifs are represented as a type of sequence alignment, which STAMP parses and converts into frequency matrices. AlignACE output.
MEME (text file output) The second or third line in MEME output files begin with the word "MEME". The "position-specific probability matrix" parts of the file are scanned for and parsed as described as above for individual MEME-format motifs. MEME output.
Weeder Weeder output files always contain the line “**** MY ADVICE ****” followed three lines later by “*** Interesting motifs (highest-ranking) seem to be”. Motifs are marked by the string “Frequency Matrix” and consist of a set of 9-column lines (beginning 3 lines after the marker), of which columns 2-5 are A C G T frequency counts. Weeder output.
MotifSampler STAMP recognizes MotifSampler matrix files (these are the .mtrx files emailed from the online version). The files always begin with the string "#INCLUSive Motif Model". PSSMs are represented as 4-column frequency matrices preceded by some description lines. MotifSampler output.
YMF STAMP recognizes both YMF command-line and web program output. The formats begin (respectively) with the lines "The best X candidates in category Y are :" or "Motif Count Zscore". YMF motifs are given one-per-line in consensus sequence format. YMF output.
ANN-Spec (command-line only) ANN-Spec command line output files begin with the line "SQI SEQUENCE_INFORMATION:". The predicted motifs (in 4xL matrix format) are prefaced with "ALR" tags. ANN-Spec output.
Consensus (command-line only) Consensus output files contain the string "THE LIST OF MATRICES FROM FINAL CYCLE--" Motifs are in 4xL format (JASPAR-like). Consensus output.
Improbizer STAMP recognizes Improbizer HTML output (as opposed to the motif file output from the command-line version of the program). Improbizer output begins with the text "Improbizer Results", and the motifs are in 4xL format, where the frequencies begin with lowercase nucleotide tags. Improbizer output.
Co-Bind Co-Bind output file's first line contain the string "# reading predefined alphabet from file". The second line contains the string "# ***** sequence information from sequence set". The matrices are in 4xL format, coming after the "ALIGNMENT_MATRIX" tag. Co-Bind output.
DME or CREAD DME and CREAD use a format that is similar to the TRANSFAC format described above. CREAD sample
NestedMICA NestedMICA's output format is the XMS (XML-like) format described above. XMS sample



Motif Trimming:
Motifs produced by many motif-finders or stored in the databases often contain a "core" region of high information-content flanked by low information-content columns at the edges. Many researchers assume that most of these flanking columns are irrelevant to the protein-DNA interaction. Whether or not this assumption is true, STAMP allows the option of stripping these low information-content edges from the input motifs. Since STAMP's motif alignment p-value calculation is dependent on the length of the compared motifs, removing the low information-content edges can help accurate alignment. STAMP allows the user to choose an information content threshold (between 0 and 1) for the purposes of excluding edge columns. The motif will not be shortened below the minimum motif length of 4 columns. For example, stripping edge columns with information content of less than 0.5 converts this motif (the JASPAR Gfi motif):
into this:
However, the user should beware that, if improperly used, this option may have the unintended consequence of removing important motif columns. For example, using a threshold of 1.0 will convert JASPAR's Nkx2-5 motif from:
into this:
Clearly, the edge columns in the original Nkx2-5 motif are somewhat informative and should not have been removed. Use the edge threshold with caution!

Similarity Matching:
For each input motif, STAMP returns a number of the closest matches in a choice of databases. Currently supported databases include:

The user may also choose to upload a dataset of their own motifs in one of the above acceptable formats for the purposes of matching the input motifs. To do so, please choose the "User-defined" option from the dropdown menu and input the motif dataset in the input box or file dialog.

Column Comparison Metrics:
STAMP allows the user to choose between a number of supported column comparison metrics to use when comparing the columns of the input matrices. The various metrics are more fully described in our publication describing STAMP. Also see the following studies for information regarding the metrics:
Pearson Correlation Coefficient: Pietrokovski S (1996) Nucleic Acids Res 24:3836-3845
Average Log Likelihood Ratio (ALLR): Wang T & Stormo GD (2003) Bioinformatics 19:2369-2380
Sum of Squared Distances: Sandelin A & Wasserman WW (2004) J Mol Biol 338:207-215
Kullback-Liebler (Relative Entropy): Roepcke S, et al. (2005) Nucleic Acids Res 33:W438-441
In our study, we found that while ALLR is the most effective in comparisons of single columns, the Pearson Correlation Coefficient and Sum of Squared Distances metrics were more effective in the context of comparing entire motifs.

Alignment Methods:
Three alignment methods are supported in STAMP. These include the local alignment Smith-Waterman strategy and the global alignment Needleman-Wunsch strategy. For these two methods, a variety of affine gap costs (where the gap extension penalty is less than the gap opening penalty) may be used. In addition, the user may choose to require local alignments where the motifs overlap significantly in the alignment, and/or where the alignment is extended past the local region of similarity to the motif edges (the latter would be a type of locally-initialized pseudo-global alignment). The third supported alignment option is a special case for ungapped Smith-Waterman (local) alignment. In this alignment option, the motif "cores" (i.e. motifs trimmed with an information content filter of 0.3) are aligned first before extending the alignment to the motif edges. This special case was found to have advantages when aligning groups of short motifs. In general, our study found that local alignments are more effective at aligning DNA motifs than global alignments.

Multiple Alignment Strategies:
STAMP offers a choice between two multiple alignment strategies. The Progressive Profile Alignment option initializes an alignment using a UPGMA tree, and builds the multiple alignment by progressively adding alignments at the nodes of the tree (starting from the leaf nodes). Iterative refinement multiple alignment takes longer, but should be more effective. Once an initial alignment is constructed, iterative refinement tries to optimize the alignment by iteratively removing a motif from the current alignment and adding it again to the remaining alignment.

Tree Algorithms:
Two tree-construction algorithms are provided. The first is the popular Unweighted Pair Group Method with Arithmetic mean (UPGMA). UPGMA is an agglomerative method that builds a tree by progressively merging the most similar nodes at each time step (beginning from the input motifs which serve as leaves). A self-organizing neural tree algorithm (SOTA) is also provided as an alternative to UPGMA. SOTA is a divisive method; it begins with a root node and two leaves and assigns motifs to the appropriate most similar node. A neural clustering approach is followed to optimize the "cluster centers" at the current leaf nodes, and then the leaf node with the most internal heterogeneity is split into two. This procedure is followed progressively until each leaf node represents a single input motif. Because of instabilities, SOTA may only be used with the ungapped Smith-Waterman alignment method at this time.

Results:
STAMP provides the results from a number of analyses of the input motif set, and the results may be downloaded either as a complete webpage or as a PDF.

Firstly, a multiple alignment of the input motifs is provided. While the multiple alignment was carried out on the motif matrices themselves, the multiple alignment is represented using consensus sequences (for implementation simplicity). The consensus alphabet used here is not the literal IUPAC consensus sequence rules (using this usually leads to consensus sequences that are degenerate beyond recognition), but rather using the following probability thresholds:
A/C/G/T is used if the appropriate single base frequency is >0.6
M/R/W/S/Y/K is used if the sum of the appropriate two bases is >0.8
N is used otherwise.
A "familial binding profile" based on the final multiple alignment is also provided as a sequence logo and as a TRANSFAC-format frequency matrix.

A tree showing the similarity between the input motifs is shown next. This tree should be useful when trying to elucidate the relationships between motifs of transcription factors from the same structural class or when trying to find degeneracy in the results of motif-finders. Next to the tree figure, the input motifs are displayed as sequence logos in the same order as the appear on the tree. Alongside the input motif logos, the best match in the chosen motif database is also shown as a sequence logo.

Finally, for each input motif, a number of the top database matches (as decided by the user) are shown in more detail. An alignment between the input motif and the match is shown as a consensus sequence. The p-value of the alignment is also shown. The p-values are calculated using the methods described by Sandelin & Wasserman. While this method aims to be independent of the motif lengths, it is not perfectly so. Therefore, it is possible to get a high score even if the motifs do not seem to be all that similar, especially if one of the motifs is large. For this reason, the p-value should not be taken as an accurate measure of the probability of two motifs being identical, but rather as a relative measure of similarity.



Please contact me (mahony(AT)mit.edu) with any further questions or suggestions.