TFBS::Matrix::PFM(3) class for raw position frequency matrix patterns


  • creating a TFBS::Matrix::PFM object manually:

    my $matrixref = [ [ 12, 3, 0, 0, 4, 0 ],
    [ 0, 0, 0, 11, 7, 0 ],
    [ 0, 9, 12, 0, 0, 0 ],
    [ 0, 0, 0, 1, 1, 12 ]
    my $pfm = TFBS::Matrix::PFM->new(-matrix => $matrixref,
    -name => "MyProfile",
    -ID => "M0001"
    # or

    my $matrixstring =
    "12 3 0 0 4 0\n0 0 0 11 7 0\n0 9 12 0 0 0\n0 0 0 1 1 12";

    my $pfm = TFBS::Matrix::PFM->new(-matrixstring => $matrixstring,
    -name => "MyProfile",
    -ID => "M0001"

  • retrieving a TFBS::Matix::PFM object from a database:

    (See documentation of individual TFBS::DB::* modules to learn how to connect to different types of pattern databases and retrieve TFBS::Matrix::* objects from them.)

        my $db_obj = TFBS::DB::JASPAR2->new
                        (-connect => ["dbi:mysql:JASPAR2:myhost",
                                      "myusername", "mypassword"]);
        my $pfm = $db_obj->get_Matrix_by_ID("M0001", "PFM");
        # or
        my $pfm = $db_obj->get_Matrix_by_name("MyProfile", "PFM");
  • retrieving list of individual TFBS::Matrix::PFM objects from a TFBS::MatrixSet object

    (See the TFBS::MatrixSet to learn how to create objects for storage and manipulation of multiple matrices.)

        my @pfm_list = $matrixset->all_patterns(-sort_by=>"name");
  • convert a raw frequency matrix to other matrix types:

        my $pwm = $pfm->to_PWM(); # convert to position weight matrix
        my $icm = $icm->to_ICM(); # convert to information con


TFBS::Matrix::PFM is a class whose instances are objects representing raw position frequency matrices (PFMs). A PFM is derived from N nucleotide patterns of fixed size, e.g. the set of sequences


will give the matrix:

    A:[ 12  3  0  0  4  0 ]
    C:[  0  0  0 11  7  0 ]
    G:[  0  9 12  0  0  0 ]
    T:[  0  0  0  1  1 12 ]

which contains the count of each nucleotide at each position in the sequence. (If you have a set of sequences as above and want to create a TFBS::Matrix::PFM object out of them, have a look at TFBS::PatternGen::SimplePFM module.)

PFMs are easily converted to other types of matrices, namely information content matrices and position weight matrices. A TFBS::Matrix::PFM object has the methods to_ICM and to_PWM which do just that, returning a TFBS::Matrix::ICM and TFBS::Matrix::PWM objects, respectively.


Please send bug reports and other comments to the author.

AUTHOR - Boris Lenhard

Boris Lenhard <[email protected]>


The rest of the documentation details each of the object methods. Internal methods are preceded with an underscore.


 Title   : new
 Usage   : my $pfm = TFBS::Matrix::PFM->new(%args)
 Function: constructor for the TFBS::Matrix::PFM object
 Returns : a new TFBS::Matrix::PFM object
 Args    : # you must specify either one of the following three:
           -matrix,      # reference to an array of arrays of integers
           -matrixstring,# a string containing four lines
                         # of tab- or space-delimited integers
           -matrixfile,  # the name of a file containing four lines
                         # of tab- or space-delimited integers
           -name,        # string, OPTIONAL
           -ID,          # string, OPTIONAL
           -class,       # string, OPTIONAL
           -tags         # an array reference, OPTIONAL
Warnings  : Warns if the matrix provided has columns with different
            sums. Columns with different sums contradict the usual
            origin of matrix data and, unless you are absolutely sure
            that column sums _should_ be different, it would be wise to
            check your matrices.


 Title   : column_sum
 Usage   : my $nr_sequences = $pfm->column_sum()
 Function: calculates the sum of elements of one column
           (the first one by default) which normally equals the
           number of sequences used to derive the PFM. 
 Returns : the sum of elements of one column (an integer)
 Args    : columnn number (starting from 1), OPTIONAL - you DO NOT
           need to specify it unless you are dealing with a matrix


 Title   : to_PWM
 Usage   : my $pwm = $pfm->to_PWM()
 Function: converts a raw frequency matrix (a TFBS::Matrix::PFM object)
           to position weight matrix. At present it assumes uniform
           background distribution of nucleotide frequencies.
 Returns : a new TFBS::Matrix::PWM object
 Args    : none; in the future releases, it should be able to accept
           a user defined background probability of the four


 Title   : to_ICM
 Usage   : my $icm = $pfm->to_ICM()
 Function: converts a raw frequency matrix (a TFBS::Matrix::PFM object)
           to information content matrix. At present it assumes uniform
           background distribution of nucleotide frequencies.
 Returns : a new TFBS::Matrix::ICM object
 Args    : -small_sample_correction # undef (default), 'schneider' or 'pseudocounts'

How a PFM is converted to ICM:

For a PFM element PFM[i,k], the probability without pseudocounts is estimated to be simply

  p[i,k] = PFM[i,k] / Z

where - Z equals the column sum of the matrix i.e. the number of motifs used to construct the PFM. - i is the column index (position in the motif) - k is the row index (a letter in the alphacer, here k is one of (A,C,G,T)

Here is how one normally calculates the pseudocount-corrected positional probability p'[i,j]:

  p'[i,k] = (PFM[i,k] + 0.25*sqrt(Z)) / (Z + sqrt(Z))

0.25 is for the flat distribution of nucleotides, and sqrt(Z) is the recommended pseudocount weight. In the general case,

  p'[i,k] = (PFM[i,k] + q[k]*B) / (Z + B)

where q[k] is the background distribution of the letter (nucleotide) k, and B an arbitrary pseudocount value or expression (for no pseudocounts B=0).

For a given position i, the deviation from random distribution in bits is calculated as (Baldi and Brunak eq. 1.9 (2ed) or 1.8 (1ed)):

- for an arbitrary alphabet of A letters:

  D[i] = log2(A) + sum_for_all_k(p[i,k]*log2(p[i,k]))

- special case for nucleotides (A=4)

  D[i] = 2 + sum_for_all_k(p[i,k]*log2(p[i,k]))

D[i] equals the information content of the position i in the motif. To calculate the entire ICM, you have to calculate the contrubution of each nucleotide at a position i to D[i], i.e.

ICM[i,k] = p'[i,k] * D[i]


 Title   : draw_logo
 Usage   : my $gd_image = $pfm->draw_logo()
 Function: draws a sequence logo; similar to the 
           method in TFBS::Matrix::ICM, but can automatically calculate
           error bars for drawing
 Returns : a GD image object (see documentation of GD module)
 Args    : many; PFM-specific options are:
           -small_sample_correction # One of 
                                    # "Schneider" (uses correction 
                                    #   described by Schneider et al.
                                    #   (Schneider t et al. (1986) J.Biol.Chem.
                                    # "pseudocounts" - standard pseudocount 
                                    #   correction,  more suitable for 
                                    #   PFMs with large r column sums
                                    # If the parameter is ommited, small
                                    # sample correction is not applied
           -draw_error_bars         # if true, adds error bars to each position
                                    # in the logo. To calculate the error bars,
                                    # it uses the -small_sample_connection
                                    # argument if explicitly set,  
                                    # or "Schneider" by default
For other args, see draw_logo entry in TFBS::Matrix::ICM documentation


 Title   : add_PFM
 Usage   : $pfm->add_PFM($another_pfm)
 Function: adds the values of $pnother_pfm matrix to $pfm
 Returns : reference to the updated $pfm object
 Args    : a TFBS::Matrix::PFM object









The above methods are common to all matrix objects. Please consult TFBS::Matrix to find out how to use them.