Scribd Logo
Upload

Welcome to Scribd!

  • 555 views

    PSSM

    Uploaded by

    rayx_1
    Position specific scoring matrices (PSSMs) are matrices created from the amino acid or nucleic acid frequencies at each position in a multiple sequence alignment. They assign higher scores to residues that appear more frequently at a given position than expected by chance. PSSMs are created by applying pseudocounts to account for residues not observed in the alignment and taking the logarithm of observed frequencies divided by background frequencies. Profile hidden Markov models (HMMs) can also represent sequence profiles and include information on consensus sequences, insertion and deletion penalties.

    Copyright:

    © All Rights Reserved
    Download as PPT, PDF, TXT or read online from Scribd

    PSSM

    Uploaded by

    rayx_1
    555 views17 pages
    Position specific scoring matrices (PSSMs) are matrices created from the amino acid or nucleic acid frequencies at each position in a multiple sequence alignment. They assign higher scores to residues that appear more frequently at a given position than expected by chance. PSSMs are created by applying pseudocounts to account for residues not observed in the alignment and taking the logarithm of observed frequencies divided by background frequencies. Profile hidden Markov models (HMMs) can also represent sequence profiles and include information on consensus sequences, insertion and deletion penalties.

    Original Description:

    Position Specific Score Matrix

    Copyright

    © © All Rights Reserved

    Available Formats

    PPT, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Position specific scoring matrices (PSSMs) are matrices created from the amino acid or nucleic acid frequencies at each position in a multiple sequence alignment. They assign higher scores to residues that appear more frequently at a given position than expected by chance. PSSMs are created by applying pseudocounts to account for residues not observed in the alignment and taking the logarithm of observed frequencies divided by background frequencies. Profile hidden Markov models (HMMs) can also represent sequence profiles and include information on consensus sequences, insertion and deletion penalties.

    Copyright:

    © All Rights Reserved
    Download as PPT, PDF, TXT or read online from Scribd
    Download as ppt, pdf, or txt
    555 views17 pages

    PSSM

    Uploaded by

    rayx_1
    Position specific scoring matrices (PSSMs) are matrices created from the amino acid or nucleic acid frequencies at each position in a multiple sequence alignment. They assign higher scores to residues that appear more frequently at a given position than expected by chance. PSSMs are created by applying pseudocounts to account for residues not observed in the alignment and taking the logarithm of observed frequencies divided by background frequencies. Profile hidden Markov models (HMMs) can also represent sequence profiles and include information on consensus sequences, insertion and deletion penalties.

    Copyright:

    © All Rights Reserved
    Download as PPT, PDF, TXT or read online from Scribd
    Download as ppt, pdf, or txt
    You are on page 1of 17

    Scoring Matrices

    Scoring matrices, PSSMs


    Position Specific Score Matrix
    (PSSM)
    A position specific scoring matrix (PSSM) is a
    matrix based on the amino acid frequencies
    (or nucleic acid frequencies) at every position
    of a multiple alignment.
    From these frequencies, the PSSM that will
    be calculated will result in a matrix that will
    assign superior scores to residues that
    appear more often than by chance at a
    certain position.
    Creating a PSSM: Example
    NTEGEWI
    NITRGEW
    NIAGECC
    Amino acid frequencies at every
    position of the alignment:
    Creating a PSSM: Example
    Amino acids that do not appear at a specific position of a multiple
    alignment must also be considered in order to model every possible
    sequence and have calculable log-odds scores. A simple procedure
    called pseudo-counts assigns minimal scores to residues that do not
    appear at a certain position of the alignment according to the
    following equation:

    Where
    Frequency is the frequency of residue i in column j (the count of
    occurances).
    pseudocount is a number higher or equal to 1.
    N is the number of sequences in the multiple alignment.
    Creating a PSSM: Example
    In this example, N = 3 and lets use pseudocount = 1:

    Score(N) at position 1 = 3/3 = 1.


    Score(I) at position 1 = 0/3 = 0.
    Readjust:
    Score(I) at position 1 -> (0+1) / (3+20) = 1/23 = 0.044.
    Score(N) at position 1 -> (3+1) / (3+20) = 4/23 = 0.174.

    The PSSM is obtained by taking the logarithm of (the values obtained above
    divided by the background frequency of the residues).

    To simplify for this example well assume that every amino acid appears equally in
    protein sequences, i.e. fi = 0.05 for every i):

    PSSM Score(I) at position 1 = log(0.044 / 0.05) = -0.061.


    PSSM Score(N) at position 1 = log(0.174 / 0.05) = 0.541.
    Creating a PSSM: Example

    The matrix assigns positive scores to residues


    that appear more often than expected by
    chance and negative scores to residues that
    appear less often than expected by chance.
    Using a PSSM
    To search for matches to a PSSM, scan along a the
    sequence using a window the length (L) of the
    PSSM.
    The matrix is slid on a sequence one residue at a
    time and the scores of the residues of every region of
    length L are added.
    NTEGEWI
    Scores that are higher than an empirically
    NITRGEW
    predetermined threshold are reported.
    NIAGECC
    Advantages of PSSM

    Weights sequence according to


    observed diversity specific to the family
    of interest
    Minimal assumptions
    Easy to compute
    Can be used in comprehensive
    evaluations.
    More sophisticated PSSMs
    From less to more complicated

    1. PSSM with pseudocounts.


    2. Giving pseudocounts less weight when more
    alignment data is available.
    3. Weight pseudocount amino acids by their
    frequency of occurrence in proteins.
    4. Instead of giving pseudocounts all the same
    value, weight them by their similarity to the
    consensus (like BLOSUM62 does) at each
    position. (PSI-BLAST method).
    5. Combine 2 & 4 (Dirichlet mixture method).
    Method 1 and
    standard
    BLOSUM62
    matrix

    Method 5

    A PSSM column with a perfectly conserved isoleucine with different


    methods used to calculate the scores.
    Using Hidden Markov models to
    describe sequence alignment profiles

    A profile HMM can represent a sequence


    alignment profile similar to how a PSSM does.

    A profile HMM includes information on the


    amino acid consensus at each position in the
    alignment like a PSSM.

    A profile HMM also has position-specific


    scores for gap insertion and extensions.
    Background: Creating HMMs

    To create an HMM to model data we need to


    determine two things:

    The structure/topology of the HMMstates


    and transitions
    The values of the parametersemission and
    transition probabilities.
    Determining the parameters is called
    training.
    A HMM structure/topology

    M=matchstate(scoretheaainthesequenceatthispositioninthe
    profile)
    I=insertion(w.r.tprofileinsertgapcharactersinprofile)
    D=deletion(w.r.tsequenceinsertgapcharactersinsequence)

    M1isfirstaaintheprofile,M2issecond,etc.
    Example HMMER parameters
    NULE 595 -1558 85 338 -294 453 -1158 (...) -21 -313 45 531 201 384
    HMM A C D E F G H (...) m->m m->i m->d i->m i->i d->m d->d b->m m->e
    1 -1084 390 -8597 -8255 -5793 -8424 -8268 (...) 1
    - -149 -500 233 43 -381 399 106 (...)
    C -1 -11642 -12684 -894 -1115 -701 -1378 -16 *
    2 -2140 -3785 -6293 -2251 3226 -2495 -727 (...) 2
    - -149 -500 233 43 -381 399 106 (...)
    C -1 -11642 -12684 -894 -1115 -701 -1378 * * (...)
    76 -2255 -5128 -302 363 -784 -2353 1398 (...) 103
    - -149 -500 233 43 -381 399 106 (...)
    E -1 -11642 -12684 -894 -1115 -701 -1378 * *
    77 -633 879 -2198 -5620 -1457 -5498 -4367 (...) 104
    - * * * * * * * (...)
    C * * * * * * * * 0
    //
    A profile HMM with match state
    probabilities shown

    AAs PATH is the consensus sequence.


    Building a profile HMM

    Pick a HMM structure/topology.


    Estimate initial parameters.
    Train the HMM by running sequences
    through it.
    Transitions that get used are given
    higher probabilities, those rarely used
    are given lower probabilities.
    Protein profile HMMs
    Better (in theory) representations than PSSMs.
    More complicated.
    Not hand-tuned by curators.
    Used in some protein profile databases:
    Pfam (http://pfam.sanger.ac.uk/)
    SMART (http://smart.embl-heidelberg.de/)
    Difficult to describe in human readable formats.

    Schuster-Bckler et al., 2004 (http://www.biomedcentral.com/1471-2105/5/7)

    You might also like