1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
|
/*
* 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 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 "kernel/LocalAlignmentStringKernel.h"
/****************/
/* The alphabet */
/****************/
#define NAA 20 /* Number of amino-acids */
#define NLET 26 /* Number of letters in the alphabet */
static const char *aaList= "ARNDCQEGHILKMFPSTWYV"; /* The list of amino acids */
/*****************/
/* SW parameters */
/*****************/
#define OPENING 12 /* Gap opening penalty */
#define EXTENSION 2 /* Gap extension penalty */
/* mutation matrix */
const int 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 */
#define BINDEX(i,j) (((i)>(j))?(j)+(((i)*(i+1))/2):(i)+(((j)*(j+1))/2))
/*********************
* Kernel parameters *
*********************/
#define 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 */
#define LOG0 -10000 /* log(0) */
#define INTSCALE 1000.0 /* critical for speed and precise computation*/
int CLocalAlignmentStringKernel::logsum_lookup[LOGSUM_TBL];
CLocalAlignmentStringKernel::CLocalAlignmentStringKernel(INT size)
: CStringKernel<CHAR>(size), initialized(false)
{
scaled_blosum=new int[sizeof(blosum)];
init_logsum();
initialize();
}
CLocalAlignmentStringKernel::CLocalAlignmentStringKernel(
CStringFeatures<CHAR>* l, CStringFeatures<CHAR>* r)
: CStringKernel<CHAR>(10), initialized(false)
{
scaled_blosum=new int[sizeof(blosum)];
init_logsum();
initialize();
init(l, r);
}
CLocalAlignmentStringKernel::~CLocalAlignmentStringKernel()
{
cleanup();
}
bool CLocalAlignmentStringKernel::init(CFeatures* l, CFeatures* r)
{
bool result = CStringKernel<CHAR>::init(l, r);
initialized = true;
return result;
}
void CLocalAlignmentStringKernel::cleanup()
{
delete[] scaled_blosum;
scaled_blosum=NULL;
free(isAA);
isAA=NULL;
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(void){
int i;
for (i = 0; i < LOGSUM_TBL; i++)
logsum_lookup[i] = (int) (INTSCALE*
(log(1.+exp( (float) -i/INTSCALE))));
}
int CLocalAlignmentStringKernel::LogSum(int p1, int p2){
int diff;
static int 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];
}
float CLocalAlignmentStringKernel::LogSum2(float p1, float 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::initialize(void)
/* Initialize all static variables. This function should be called once before computing the first pair HMM score */
{
register int i;
/* Initialization of the array which gives the position of each amino-acid in the set of amino-acid */
if ((aaIndex=(int *)calloc(NLET,sizeof(int))) == NULL)
SG_ERROR("run out o memory");
for (i=0;i<NAA;i++)
aaIndex[aaList[i]-'A']=i;
/* Initialization of the array which indicates whether a char is an amino-acid */
if ((isAA=(int *)calloc(256,sizeof(int))) == NULL)
SG_ERROR("run out of memory");
for (i=0;i<NAA;i++)
isAA[(int)aaList[i]]=1;
/* Scale the blossum matrix */
for (i=0 ; i<NAA*(NAA+1)/2; i++)
scaled_blosum[i] = (int) floor(blosum[i]*SCALING*INTSCALE);
/* Scale of gap penalties */
opening = (int) floor(OPENING * SCALING*INTSCALE);
extension = (int) floor(EXTENSION * SCALING*INTSCALE);
}
DREAL CLocalAlignmentStringKernel::LAkernelcompute(int* aaX, int* aaY, /* Implementation of the
convolution kernel which generalizes the Smith-Waterman algorithm */
/* the two amino-acid sequences (as sequences of indexes in [0..NAA-1] indicating the
* position of the amino-acid in the variable 'aaList') */
int nX, int nY /* the lengths of both sequences */
)
{
register int
i,j, /* loop indexes */
cur, old, /* to indicate the array to use (0 or 1) */
curpos, frompos; /* position in an array */
int
*logX, /* arrays to store the log-values of each state */
*logY,
*logM,
*logX2,
*logY2,
aux , aux2;/* , aux3 , aux4 , aux5;*/
int
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=new int[2*cl];
logX=new int[2*cl];
logY=new int[2*cl];
logX2=new int[2*cl];
logY2=new int[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( - opening + logM[frompos] , - 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( - opening + logM[frompos] , - extension + logY[frompos] );
logY[curpos] = LOGP( aux , - 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 */
delete[] logM;
delete[] logX;
delete[] logY;
delete[] logX2;
delete[] logY2;
/* Return the logarithm of the kernel */
return (float)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" */
DREAL CLocalAlignmentStringKernel::compute(INT idx_x, INT idx_y)
{
int *aax,*aay; /* to convert x and y into sequences of amino-acid indexes */
int lx=0,ly=0; /* lengths of x and y */
int i,j;
/* If necessary, initialize static variables */
if (isAA == NULL)
initialize();
CHAR* x=((CStringFeatures<CHAR>*) lhs)->get_feature_vector(idx_x, lx);
CHAR* y=((CStringFeatures<CHAR>*) rhs)->get_feature_vector(idx_y, ly);
ASSERT(x && y);
if ((lx<1) || (ly<1))
SG_ERROR("empty chain");
/* Create aax and aay */
if ((aax=(int *)calloc(lx,sizeof(int))) == NULL)
SG_ERROR("run out of memory");
if ((aay=(int *)calloc(ly,sizeof(int))) == NULL)
SG_ERROR("run out of memory");
/* 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 */
DREAL result=LAkernelcompute(aax,aay,lx,ly);
/* Release memory */
free(aax);
free(aay);
return result;
}
|