LocalAlignmentStringKernel.cpp

Go to the documentation of this file.

/*
 * Compute the local alignment kernel
 *
 * Largely based on LAkernel.c (version 0.3)
 *
 * Copyright 2003 Jean-Philippe Vert
 * Copyright 2005 Jean-Philippe Vert, Hiroto Saigo
 *
 * Shogun specific adjustments Written (W) 2007-2008,2010 Soeren Sonnenburg
 *
 * Reference:
 * H. Saigo, J.-P. Vert, T. Akutsu and N. Ueda, "Protein homology
 * detection using string alignment kernels", Bioinformatics,
 * vol.20, p.1682-1689, 2004.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 3 of the License, or
 * (at your option) any later version.
 */

#include <stdlib.h>
#include <stdio.h>
#include <math.h>
#include <ctype.h>
#include <string.h>
#include <shogun/kernel/string/LocalAlignmentStringKernel.h>

using namespace shogun;

/****************/
/* The alphabet */
/****************/

const int32_t NAA=20;                                  /* Number of amino-acids */
const int32_t NLET=26;                                 /* Number of letters in the alphabet */
const char* CLocalAlignmentStringKernel::aaList="ARNDCQEGHILKMFPSTWYV";    /* The list of amino acids */

/*****************/
/* SW parameters */
/*****************/

/* mutation matrix */
const int32_t CLocalAlignmentStringKernel::blosum[] = {
  6,
 -2,   8,
 -2,  -1,   9,
  -3,  -2,   2,   9,
  -1,  -5,  -4,  -5,  13,
  -1,   1,   0,   0,  -4,   8,
  -1,   0,   0,   2,  -5,   3,   7,
  0,  -3,  -1,  -2,  -4,  -3,  -3,   8,
  -2,   0,   1,  -2,  -4,   1,   0,  -3,  11,
  -2,  -5,  -5,  -5,  -2,  -4,  -5,  -6,  -5,   6,
  -2,  -3,  -5,  -5,  -2,  -3,  -4,  -5,  -4,   2,   6,
  -1,   3,   0,  -1,  -5,   2,   1,  -2,  -1,  -4,  -4,   7,
  -1,  -2,  -3,  -5,  -2,  -1,  -3,  -4,  -2,   2,   3,  -2,   8,
  -3,  -4,  -5,  -5,  -4,  -5,  -5,  -5,  -2,   0,   1,  -5,   0,   9,
  -1,  -3,  -3,  -2,  -4,  -2,  -2,  -3,  -3,  -4,  -4,  -2,  -4,  -5,  11,
  2,  -1,   1,   0,  -1,   0,   0,   0,  -1,  -4,  -4,   0,  -2,  -4,  -1,   6,
  0,  -2,   0,  -2,  -1,  -1,  -1,  -2,  -3,  -1,  -2,  -1,  -1,  -3,  -2,   2,   7,
  -4,  -4,  -6,  -6,  -3,  -3,  -4,  -4,  -4,  -4,  -2,  -4,  -2,   1,  -6,  -4,  -4,  16,
  -3,  -3,  -3,  -5,  -4,  -2,  -3,  -5,   3,  -2,  -2,  -3,  -1,   4,  -4,  -3,  -2,   3,  10,
  0,  -4,  -4,  -5,  -1,  -3,  -4,  -5,  -5,   4,   1,  -3,   1,  -1,  -4,  -2,   0,  -4,  -2,   6};

/* Index corresponding to the (i,j) entry (i,j=0..19) in the blosum matrix */
static int32_t BINDEX(int32_t i, int32_t j)
{
    return (((i)>(j))?(j)+(((i)*(i+1))/2):(i)+(((j)*(j+1))/2));
}

/*********************
 * Kernel parameters *
 *********************/

const float64_t SCALING=0.1;           /* Factor to scale all SW parameters */

/* If you want to compute the sum over all local alignments (to get a valid kernel), uncomment the following line : */
/* If x=log(a) and y=log(b), compute log(a+b) : */
/*
#define LOGP(x,y) (((x)>(y))?(x)+log1p(exp((y)-(x))):(y)+log1p(exp((x)-(y))))
*/

#define LOGP(x,y) LogSum(x,y)

/* OR if you want to compute the score of the best local alignment (to get the SW score by Viterbi), uncomment the following line : */
/*
#define LOGP(x,y) (((x)>(y))?(x):(y))
*/

/* Usefule constants */
const float64_t LOG0=-10000;          /* log(0) */
const float64_t INTSCALE=1000.0;      /* critical for speed and precise computation*/

int32_t CLocalAlignmentStringKernel::logsum_lookup[LOGSUM_TBL];

CLocalAlignmentStringKernel::CLocalAlignmentStringKernel(int32_t size)
: CStringKernel<char>(size)
{
    SG_UNSTABLE("LocalAlignmentStringKernel");
    init();
    init_static_variables();
}

CLocalAlignmentStringKernel::CLocalAlignmentStringKernel(
    CStringFeatures<char>* l, CStringFeatures<char>* r,
    float64_t opening, float64_t extension)
: CStringKernel<char>()
{
    SG_UNSTABLE("LocalAlignmentStringKernel");
    init();
    m_opening=opening;
    m_extension=extension;
    init_static_variables();
    init(l, r);
}

CLocalAlignmentStringKernel::~CLocalAlignmentStringKernel()
{
    cleanup();
}

bool CLocalAlignmentStringKernel::init(CFeatures* l, CFeatures* r)
{
    CStringKernel<char>::init(l, r);
    initialized=true;
    return init_normalizer();
}

void CLocalAlignmentStringKernel::cleanup()
{
    SG_FREE(scaled_blosum);
    scaled_blosum=NULL;

    SG_FREE(isAA);
    isAA=NULL;
    SG_FREE(aaIndex);
    aaIndex=NULL;

    CKernel::cleanup();
}

/* LogSum - default log funciotion. fast, but not exact */
/* LogSum2 - precise, but slow. Note that these two functions need different figure types  */

void CLocalAlignmentStringKernel::init_logsum(){
    int32_t i;
    for (i=0; i<LOGSUM_TBL; i++)
        logsum_lookup[i]=int32_t(INTSCALE*
                (log(1.+exp((float32_t)-i/INTSCALE))));
}

int32_t CLocalAlignmentStringKernel::LogSum(int32_t p1, int32_t p2)
{
    int32_t diff;
    static int32_t firsttime=1;

    if (firsttime)
    {
        init_logsum();
        firsttime =0;
    }

    diff=p1-p2;
    if (diff>=LOGSUM_TBL)
        return p1;
    else if (diff<=-LOGSUM_TBL)
        return p2;
    else if (diff>0)
        return p1+logsum_lookup[diff];
    else
        return p2+logsum_lookup[-diff];
}


float32_t CLocalAlignmentStringKernel::LogSum2(float32_t p1, float32_t p2)
{
    if (p1>p2)
        return (p1-p2>50.) ? p1 : p1+log(1.+exp(p2-p1));
    else
        return (p2-p1>50.) ? p2 : p2+log(1.+exp(p1-p2));
}


void CLocalAlignmentStringKernel::init_static_variables()
     /* Initialize all static variables. This function should be called once before computing the first pair HMM score */
{
    register int32_t i;

    /* Initialization of the array which gives the position of each amino-acid in the set of amino-acid */
    aaIndex = SG_CALLOC(int32_t, NLET);
    for (i=0;i<NAA;i++)
        aaIndex[aaList[i]-'A']=i;

    /* Initialization of the array which indicates whether a char is an amino-acid */
    isAA = SG_CALLOC(int32_t, 256);
    for (i=0;i<NAA;i++)
        isAA[(int32_t)aaList[i]]=1;

    /* Scale the blossum matrix */
    for (i=0 ; i<NAA*(NAA+1)/2; i++)
        scaled_blosum[i]=(int32_t)floor(blosum[i]*SCALING*INTSCALE);


    /* Scale of gap penalties */
    m_opening=(int32_t)floor(m_opening*SCALING*INTSCALE);
    m_extension=(int32_t)floor(m_extension*SCALING*INTSCALE);
}



/* Implementation of the
 * convolution kernel which generalizes the Smith-Waterman algorithm
 */
float64_t CLocalAlignmentStringKernel::LAkernelcompute(
    int32_t* aaX, int32_t* aaY, /* the two amino-acid sequences (as sequences of indexes in [0..NAA-1] indicating the position of the amino-acid in the variable 'aaList') */
    int32_t nX, int32_t nY /* the lengths of both sequences */)
{
    register int32_t
    i,j,                /* loop indexes */
    cur, old,           /* to indicate the array to use (0 or 1) */
    curpos, frompos;    /* position in an array */

    int32_t
    *logX,           /* arrays to store the log-values of each state */
    *logY,
    *logM,
    *logX2,
    *logY2;

    int32_t aux, aux2;/* , aux3 , aux4 , aux5;*/
    int32_t cl;/* length of a column for the dynamic programming */

    /*
    printf("now computing pairHMM between %d and %d:\n",nX,nY);
    for (i=0;i<nX;printf("%d ",aaX[i++]));
    printf("\n and \n");
    for (i=0;i<nY;printf("%d ",aaY[i++]));
    printf("\n");
    */

    /* Initialization of the arrays */
    /* Each array stores two successive columns of the (nX+1)x(nY+1) table used in dynamic programming */
    cl=nY+1;           /* each column stores the positions in the aaY sequence, plus a position at zero */

    logM=SG_CALLOC(int32_t, 2*cl);
    logX=SG_CALLOC(int32_t, 2*cl);
    logY=SG_CALLOC(int32_t, 2*cl);
    logX2=SG_CALLOC(int32_t, 2*cl);
    logY2=SG_CALLOC(int32_t, 2*cl);

    /************************************************/
    /* First iteration : initialization of column 0 */
    /************************************************/
    /* The log=proabilities of each state are initialized for the first column (x=0,y=0..nY) */

    for (j=0; j<cl; j++) {
        logM[j]=LOG0;
        logX[j]=LOG0;
        logY[j]=LOG0;
        logX2[j]=LOG0;
        logY2[j]=LOG0;
    }

    /* Update column order */
    cur=1;      /* Indexes [0..cl-1] are used to process the next column */
    old=0;      /* Indexes [cl..2*cl-1] were used for column 0 */


    /************************************************/
    /* Next iterations : processing columns 1 .. nX */
    /************************************************/

    /* Main loop to vary the position in aaX : i=1..nX */
    for (i=1; i<=nX; i++) {

        /* Special update for positions (i=1..nX,j=0) */
        curpos=cur*cl;                  /* index of the state (i,0) */
        logM[curpos]=LOG0;
        logX[curpos]=LOG0;
        logY[curpos]=LOG0;
        logX2[curpos]=LOG0;
        logY2[curpos]=LOG0;

        /* Secondary loop to vary the position in aaY : j=1..nY */
        for (j=1; j<=nY; j++) {

            curpos=cur*cl+j;            /* index of the state (i,j) */

            /* Update for states which emit X only */
            /***************************************/

            frompos=old*cl+j;            /* index of the state (i-1,j) */

            /* State RX */
            logX[curpos]=LOGP(-m_opening+logM[frompos], -m_extension+logX[frompos]);
            /*      printf("%.5f\n",logX[curpos]);*/
            /*      printf("%.5f\n",logX_B[curpos]);*/
            /* State RX2 */
            logX2[curpos]=LOGP(logM[frompos], logX2[frompos]);

            /* Update for states which emit Y only */
            /***************************************/

            frompos=cur*cl+j-1;          /* index of the state (i,j-1) */

            /* State RY */
            aux=LOGP(-m_opening+logM[frompos], -m_extension+logY[frompos]);
            logY[curpos] = LOGP(aux, -m_opening+logX[frompos]);

            /* State RY2 */
            aux=LOGP(logM[frompos], logY2[frompos]);
            logY2[curpos]=LOGP(aux, logX2[frompos]);

            /* Update for states which emit X and Y */
            /****************************************/

            frompos=old*cl+j-1;          /* index of the state (i-1,j-1) */

            aux=LOGP(logX[frompos], logY[frompos]);
            aux2=LOGP(0, logM[frompos]);
            logM[curpos]=LOGP(aux, aux2)+scaled_blosum[BINDEX(aaX[i-1], aaY[j-1])];

            /*
            printf("i=%d , j=%d\nM=%.5f\nX=%.5f\nY=%.5f\nX2=%.5f\nY2=%.5f\n",i,j,logM[curpos],logX[curpos],logY[curpos],logX2[curpos],logY2[curpos]);
            */

        }  /* end of j=1:nY loop */


        /* Update the culumn order */
        cur=1-cur;
        old=1-old;

    }  /* end of j=1:nX loop */


    /* Termination */
    /***************/

    curpos=old*cl+nY;                /* index of the state (nX,nY) */
    aux=LOGP(logX2[curpos], logY2[curpos]);
    aux2=LOGP(0, logM[curpos]);
    /*  kernel_value = LOGP( aux , aux2 );*/

    /* Memory release */
    SG_FREE(logM);
    SG_FREE(logX);
    SG_FREE(logY);
    SG_FREE(logX2);
    SG_FREE(logY2);

    /* Return the logarithm of the kernel */
    return (float32_t)LOGP(aux,aux2)/INTSCALE;
}

/********************/
/* Public functions */
/********************/


/* Return the log-probability of two sequences x and y under a pair HMM model */
/* x and y are strings of aminoacid letters, e.g., "AABRS" */
float64_t CLocalAlignmentStringKernel::compute(int32_t idx_x, int32_t idx_y)
{
    int32_t *aax, *aay;  /* to convert x and y into sequences of amino-acid indexes */
    int32_t lx=0, ly=0;       /* lengths of x and y */
    int32_t i, j;

    bool free_x, free_y;
    char* x=((CStringFeatures<char>*) lhs)->get_feature_vector(idx_x, lx, free_x);
    char* y=((CStringFeatures<char>*) rhs)->get_feature_vector(idx_y, ly, free_y);
    ASSERT(x && y);

    if ( (lx<1) || (ly<1) )
        SG_ERROR("empty chain");

    /* Create aax and aay */

    aax = SG_CALLOC(int32_t, lx);
    aay = SG_CALLOC(int32_t, ly);

    /* Extract the characters corresponding to aminoacids and keep their indexes */

    j=0;
    for (i=0; i<lx; i++)
        if (isAA[toupper(x[i])])
            aax[j++]=aaIndex[toupper(x[i])-'A'];
    lx=j;
    j=0;
    for (i=0; i<ly; i++)
        if (isAA[toupper(y[i])])
            aay[j++]=aaIndex[toupper(y[i])-'A'];
    ly=j;


    /* Compute the pair HMM score */
    float64_t result=LAkernelcompute(aax, aay, lx, ly);

    /* Release memory */
    SG_FREE(aax);
    SG_FREE(aay);

    ((CStringFeatures<char>*)lhs)->free_feature_vector(x, idx_x, free_x);
    ((CStringFeatures<char>*)rhs)->free_feature_vector(y, idx_y, free_y);

    return result;
}

void CLocalAlignmentStringKernel::init()
{
    initialized=false;
    isAA=NULL;
    aaIndex=NULL;

    m_opening=10;
    m_extension=2;

    scaled_blosum=SG_CALLOC(int32_t, sizeof(blosum));
    init_logsum();

    SG_ADD(&initialized, "initialized", "If kernel is initalized.",
        MS_NOT_AVAILABLE);
    SG_ADD(&m_opening, "opening", "Opening gap opening penalty.", MS_AVAILABLE);
    SG_ADD(&m_extension, "extension", "Extension gap extension penalty.",
        MS_AVAILABLE);
}