Name

convert-matrix

2004 by

Description

Performs inter-conversions between various formats of position-specific scoring matrices (PSSM).

The program also performs a statistical analysis of the original matrix to provide different position-specific scores (weight, frequencies, information contents), general statistics (E-value, total information content), and synthetic descriptions (consensus).

PSSM can be used to represent the binding specificity of a transcription factor or the conserved residues of a protein domain.

Each row of the matrix corresponds to one residue (nucleotide or amino-acid depending on the sequence type). Each column corresponds to one position in the alignment. The value within each cell represents the frequency of each residue at each position.

INPUT/OUTPUT FORMATS

Some formats are supported only for input, others for output. There are more formats accepted for input, because the general use of this program is to convert a PSSM obtained from a database (e.g. TRANSFAC) or a pattern-discovery program (e.g. consensus, gibbs, meme, MotifSampler, ...) and obtain a matrix either for scanning (with matrix-scan) or for computing statistical parameters (see the return fields below).

• tab (input/output)
tab-delimited file. One row per residue, one column per position. The first column of each row indicates the residue, the following columns give the frequency of that residue at the corresponding position of the matrix.

The tab format accepts a user-specific set of return fields (option -return), proviging different statistics on the matrix (counts, frequencies, weights, information, other parameters: see description below).

     
; MET4 matrix, from Gonze et al. (2005). Bioinformatics 21, 3490-500.
A |   7   9   0   0  16   0   1   0   0  11   6   9   6   1   8
C |   5   1   4  16   0  15   0   0   0   3   5   5   0   2   0
G |   4   4   1   0   0   0  15   0  16   0   3   0   0   2   0
T |   0   2  11   0   0   1   0  16   0   2   2   2  10  11   8
//

• patser (output)
This format can be used as input to scan sequences with patser, the pattern-matching program developed by Jerry Hertz.

This is actually the same format as tab (described above), but the only return field is the count matrix.

• assembly (input)
Output file from the program pattern-assembly. One assembly file can contain zero, one or several assemblies. Each assembly is converted to a position-specific scoring matrix by taking, for each residue at each position, the score of the most significant pattern (oligonucleotide) containing that residue in this position of the assembly.
• consensus (input/output)
Output file from consensus, the pattern-discovery program developed by Jerry Hertz (Hertz et al., Comput Appl Biosci, 1990:6, 81-92). This file contains one or several matrices, + additional information on the parameters used for pattern discovery (e.g. prior residue frequencies).
• gibbs (input)
Output file from gibbs, the pattern-discovery program developed by Andrew Neuwald (Lawrence et al. Science, 1993: 262, 208-214; Neuwald, et al. Protein Sci, 1995: 4, 1618-1632)
• meme (input)
Output file from MEME, the pattern-discovery program developed by tim Bailey.This file contains one or several matrices, + additional information on the parameters used for pattern discovery (e.g. prior residue frequencies).
• MotifSampler (input/output)
Output file from MotifSampler, the pattern-discovery program developed by Gert Thijs (Thijs et al. Bioinformatics, 2001:17, 1113-1122).
• TRANSFAC (input/output)
Format used in the TRANSFAC database; (http://www.gene-regulation.com/pub/databases.html)
• cb (input)
Cluster-Buster output file (usual extention .cb). The header line starts with a > (like in fasta format). The matrix is then printed "vertically" on the following lines: each column corresponds to one residue, and each row to a position in the alignment.
• feature (input)
Output file from convert-features. This format allows to obtain a PSSM from a list of (supposedly pre-aligned) sites. These sites can themselves have been collected by scanning sequences with a matrix (matrix-scan) or by searching string-based patterns in a sequence (dna-pattern). Converting features to matrices can for example be useful for iterative refinment of a matrix (colecting sites from a matrix, and building a matrix from those sites). Another application is to detect oligomers or dyads in a sequence set, and build a matrix from these.
• clustal (input)
The popular multiple alignemnt program clustalw.

RETURN FIELDS FOR THE TAB-DELIMITED OUTPUT FORMAT

counts


Each cell of the matrix indicates the number of occurrences of the residue at a given position of the alignment.

profile


The matrix is printed vertically (each matrix column becomes a row in the output text). Additional parameters (consensus, information) are indicated besides each position, and a histogram is drawed.

crude frequencies


Relative frequencies are calculated as the counts of residues divided by the total count of the column.

Fij=Cij/SUMi(Cij)

where

Cij


is the absolute frequency (counts) of residue i at position j of the alignment

Fij


is the relative frequency of residue i at position j of the alignment

frequencies corrected with pseudo-weights


Relative frequencies can be corrected by a pseudo-weight (b) to reduce the bias due to the small number of observations.

F''ij=Cij+b*Pi/[SUMi(Cij)+b]

where

Pi


is the prior frequency for residue i

b


is the pseudo-weight, which is ``shared'' between residues according to their prior frequencies.

weights


Weights are calculated according to the formula from Hertz (1999), as the natural logarithm of the ratio between the relative frequency (corrected for pseudo-weights) and the prior residue probability.

Wij=ln(F''ij/Pi)

information


The crude information content is calculated according to the formula from Hertz (1999).

Iij = Fij*ln(Fij/Pi)

In addition, we calculate a ``corrected'' information content which takes pseudo-weights into account.

I''ij = F''ij*ln(F''ij/Pi)

P-value


The P-value indicates the probability to observe at least Cij occurrences of a residue at a given position of the matrix. It is calculated with the binomial formula:

    k=C.j    C.j!      k      Cij-k
Pij= SUM  ---------- Pi (1-Pi)
    k=Cij k!(C.j-k)!

where

Cij


is the number of occurrences of residue i at position j of the matrix.

C.j


is the sum of all residue occurrences at position j of the matrix.

Pi


is the prior probability of residue i.

parameters


Returns a series of parameters associated to the matrix. The list of parameters to be exported depends on the input formats (each pattern discovery program returns specific parameters, which are more or less related to each others but not identical).

Some additional parameters are optionally calculated

consensus


The degenerate consensus is calculated by collecting, at each position, the list of residues with a positive weight. Contrarily to most applications, this consensus is thus weighted by prior residue frequencies: a residue with a high frequency might not be represented in the consensus if this frequency does not significantly exceed the expected frequency. Uppercases are used to highlight weights >= 1.

The consensus is exported as regular expression, and with the IUPAC code for ambiguous nucleotides (http://www.chem.qmw.ac.uk/iupac/misc/naseq.html).

        A                       (Adenine) 
        C                       (Cytosine)
        G                       (Guanine)
        T                       (Thymine)
        R       = A or G        (puRines)
        Y       = C or T        (pYrimidines)
        W       = A or T        (Weak hydrogen bonding)
        S       = G or C        (Strong hydrogen bonding)
        M       = A or C        (aMino group at common position)
        K       = G or T        (Keto group at common position)
        H       = A, C or T     (not G)
        B       = G, C or T     (not A)
        V       = G, A, C       (not T)
        D       = G, A or T     (not C)
        N       = G, A, C or T  (aNy)

The strict consensus indicates, at each position, the residue with the highest positive weight.

information


The total information is calculated by summing the information content of all the cells of the matrix. This parameters is already returned by the program consensus (Hertz), but not by other programs.

logo


Sequence logo, a visual representation of the motif, where each column of the matrix is represented as a stack of letters whose size is proportional to the corresponding residue frequency. The total height of each column is proportional to its information content.

Sequence logo are generated using the freeware program Weblogo.

For information, contact