double vector_similarity(void) {
double v1[4],v2[4],sum,dist;
double *d;
int numRandom;
register int i,j;
numRandom=100000;
d=alloc_double(numRandom);
for (i=0; i<numRandom; i++) {
for (j=0; j<4; j++) {
v1[j]=genrand();
v2[j]=genrand();
}
sum=0; for (j=0; j<4; j++) sum+=v1[j];
if (sum!=0) { for (j=0; j<4; j++) v1[j]/=sum; }
else { for (j=0; j<4; j++) v1[j]=0.25; }
sum=0; for (j=0; j<4; j++) sum+=v2[j];
if (sum!=0) { for (j=0; j<4; j++) v2[j]/=sum; }
else { for (j=0; j<4; j++) v2[j]=0.25; }
d[i]=0; for (j=0; j<4; j++) { d[i] +=fabs(v1[j]-v2[j]); }
}
// sort in increasing order
sort_double2(d,numRandom);
dist=d[(int)(numRandom*SIMILARITY_ALPHA)];
if (d) { free(d); d=NULL; }
return (dist);
}
double *base_frequency(int numSeq,char **seq,int *seqLen) {
register int i,j;
int bcount[4];
int sum;
double *freq;
freq=alloc_double(4);
for (j=0; j<4; j++) bcount[j]=0;
for (i=0; i<numSeq; i++) {
for (j=0; j<seqLen[i]; j++) {
switch (seq[i][j]) {
case 'a': (bcount[0])++; break;
case 'c': (bcount[1])++; break;
case 'g': (bcount[2])++; break;
case 't': (bcount[3])++; break;
default: break;
}
}
}
sum=0; for (j=0; j<4; j++) sum +=bcount[j];
for (j=0; j<4; j++) freq[j]=(double)bcount[j]/(double)sum;
for (j=0; j<4; j++) freq[j]=(freq[j]+PSEUDO_COUNT)/(1.0+PSEUDO_COUNT*4.0);
freq[0]=(freq[0]+freq[3])/2.0; freq[3]=freq[0];
freq[1]=(freq[1]+freq[2])/2.0; freq[2]=freq[1];
return (freq);
}
void roulett_wheel_rank(Fitness *fitness,int populationSize,Wheel *wheel) {
register int i;
int sum;
double *weight;
weight=alloc_double(populationSize);
sum=0;
for (i=1; i<populationSize+1; i++) sum += i;
for (i=0; i<populationSize; i++) weight[i]=(double)(populationSize-i)/(double)sum;
wheel[0].start=0.0;
wheel[0].end =(double)populationSize*weight[0];
wheel[0].index=fitness[0].index;
for (i=1; i<populationSize; i++) {
wheel[i].start=wheel[i-1].end;
wheel[i].end=(double)populationSize*weight[i]+wheel[i].start;
wheel[i].index=fitness[i].index;
}
//for (i=0; i<populationSize; i++) {
// printf("%4d\t%8.3f\t%3d\t%8.6f\t%8.6f\n",
// fitness[i].index,fitness[i].value,wheel[i].index,wheel[i].start,wheel[i].end);
//}
if (weight) {
free(weight);
weight=NULL;
}
}
Words *alloc_word(int numCategory,int maxSize) {
int i;
Words *tmp=NULL;
tmp=(Words *)calloc(numCategory,sizeof(Words));
if (!tmp) {
error("calloc failed for Words.\n");
/*Rprintf("calloc failed for Words.\n"); exit(0); */
}
for (i=0; i<numCategory; i++) {
tmp[i].s1=alloc_char_char(maxSize,10);
tmp[i].prob_sta=alloc_double(maxSize);
tmp[i].prob_end=alloc_double(maxSize);
}
return (tmp);
}
// this subroutine assigns probabilities that are proportional to fitness scores
// Note that this minimizes rather than maximizes
// The minimum is assigned the largest probability
void roulett_wheel_fitness(Fitness *fitness,int populationSize,Wheel *wheel) {
register int i;
double totalScore,worstScore,range;
double *scaledScore;
worstScore=fitness[populationSize-1].value;
range=worstScore-fitness[0].value;
// for (i=0; i<populationSize; i++) printf("%d\t%5.3f\n",fitness[i].index,fitness[i].value);
if ((range<0.0001) ||
fitness[1].value==DUMMY_FITNESS ||
fabs(fitness[1].value-fitness[populationSize-1].value)<0.0001) {
printf("GA converged ...\n");
for (i=0; i<populationSize; i++) {
wheel[i].index=fitness[i].index;
wheel[i].start=i;
wheel[i].end=i+1;
}
}
else {
scaledScore=alloc_double(populationSize);
totalScore=0;
for (i=0; i<populationSize; i++) {
// range scale
// make sure the minimum is assigned the largest area
scaledScore[i]=1.0-(fitness[i].value-fitness[0].value)/range;
totalScore += scaledScore[i];
}
for (i=0; i<populationSize; i++) scaledScore[i] /= totalScore;
//for (i=0; i<populationSize; i++) printf("%6.5f\n",scaledScore[i]); exit(0);
wheel[0].start=0;
wheel[0].end =(double)populationSize*scaledScore[0];
wheel[0].index=fitness[0].index;
//if (fitness[0].value<0) wheel[0].I_evalue='0';
//else wheel[0].I_evalue='1';
for (i=1; i<populationSize; i++) {
wheel[i].start=wheel[i-1].end;
wheel[i].end=(double)populationSize*scaledScore[i]+wheel[i].start;
wheel[i].index=fitness[i].index;
//if (fitness[i].value<0) wheel[i].I_evalue='0';
//else wheel[i].I_evalue='1';
}
if (scaledScore) {
free(scaledScore);
scaledScore = NULL;
}
}
//for (i=0; i<populationSize; i++)
// printf("%4d\t%5.3f\t%3d\t%c\t%5.3f\t%5.3f\n",
// fitness[i].index,fitness[i].value,wheel[i].index,wheel[i].I_evalue,wheel[i].start,wheel[i].end);
}
extern "C" void
invoke_copy_to_stack(uint64_t* stk, uint64_t *end,
uint32_t paramCount, nsXPTCVariant* s)
{
uint64_t *ireg_args = stk;
uint64_t *ireg_end = ireg_args + 8;
double *freg_args = (double *)ireg_end;
double *freg_end = freg_args + 8;
uint64_t *stack_args = (uint64_t *)freg_end;
// leave room for 'that' argument in x0
++ireg_args;
for (uint32_t i = 0; i < paramCount; i++, s++) {
if (s->IsPtrData()) {
alloc_word(ireg_args, stack_args, ireg_end, (uint64_t)s->ptr);
continue;
}
// According to the ABI, integral types that are smaller than 8 bytes
// are to be passed in 8-byte registers or 8-byte stack slots.
switch (s->type) {
case nsXPTType::T_FLOAT:
alloc_float(freg_args, stack_args, freg_end, s->val.f);
break;
case nsXPTType::T_DOUBLE:
alloc_double(freg_args, stack_args, freg_end, s->val.d);
break;
case nsXPTType::T_I8: alloc_word(ireg_args, stk, end, s->val.i8); break;
case nsXPTType::T_I16: alloc_word(ireg_args, stk, end, s->val.i16); break;
case nsXPTType::T_I32: alloc_word(ireg_args, stk, end, s->val.i32); break;
case nsXPTType::T_I64: alloc_word(ireg_args, stk, end, s->val.i64); break;
case nsXPTType::T_U8: alloc_word(ireg_args, stk, end, s->val.u8); break;
case nsXPTType::T_U16: alloc_word(ireg_args, stk, end, s->val.u16); break;
case nsXPTType::T_U32: alloc_word(ireg_args, stk, end, s->val.u32); break;
case nsXPTType::T_U64: alloc_word(ireg_args, stk, end, s->val.u64); break;
case nsXPTType::T_BOOL: alloc_word(ireg_args, stk, end, s->val.b); break;
case nsXPTType::T_CHAR: alloc_word(ireg_args, stk, end, s->val.c); break;
case nsXPTType::T_WCHAR: alloc_word(ireg_args, stk, end, s->val.wc); break;
default:
// all the others are plain pointer types
alloc_word(ireg_args, stack_args, ireg_end,
reinterpret_cast<uint64_t>(s->val.p));
break;
}
}
}
void mutation(Chrs **dyad,int numWordGroup,Words *word,int minSpaceWidth,int maxSpaceWidth,
Wheel *wheel,int populationSize,Fitness *fitness,char *uniqMotif,double *maxpFactor,double maxpMutationRate) {
register int i,j;
int popuCn; // index of the number of dyads in population
int whichToMutate; // the spaced dyad for mutation
int kmerGroup; // pick a k-mer group (tetramer,pentamer,hexamer) in which a new word will be selected
int whichPartDyad; // pick a component of a spaced dyad for mutation
int spaceWidth;
double *tmpmaxpFactor;
double rand,newFactor;
Chrs **tmpDyad;
tmpDyad=alloc_chrs(populationSize,4);
tmpmaxpFactor=alloc_double(populationSize);
popuCn=0;
for (i=0; i<populationSize; i++) {
if (uniqMotif[i]=='1') {
for (j=0; j<3; j++) {
tmpDyad[popuCn][j].wordID =dyad[fitness[i].index][j].wordID;
tmpDyad[popuCn][j].wordGroup=dyad[fitness[i].index][j].wordGroup;
}
tmpmaxpFactor[popuCn]=maxpFactor[fitness[i].index];
popuCn++;
}
}
// printf("popuCn=%d\n",popuCn);
while (popuCn<populationSize) {
// sample with replacement with probability proportional to fitness score
rand=(double)populationSize*runif(0,1);
whichToMutate=0;
for (j=0; j<populationSize; j++) {
if (rand>=wheel[j].start && rand<=wheel[j].end) {
whichToMutate=wheel[j].index; break;
}
}
tmpmaxpFactor[popuCn]=maxpFactor[whichToMutate];
for (j=0; j<3; j++) {
tmpDyad[popuCn][j].wordID =dyad[whichToMutate][j].wordID;
tmpDyad[popuCn][j].wordGroup=dyad[whichToMutate][j].wordGroup;
}
if (runif(0,1)<maxpMutationRate) {
// find a different maxp factor, keep dyad unchanged
do {
newFactor=MAXP_BASE+MAXP_FACTOR*((int)(MAXP_SCALE*runif(0,1)));
} while (fabs(newFactor-tmpmaxpFactor[popuCn])<0.001);
tmpmaxpFactor[popuCn]=newFactor;
}
else {
// change one component of the dyad, first find out how many components there are
if (maxSpaceWidth==0) {
whichPartDyad=(int)(2*runif(0,1)); // change w1
if (whichPartDyad==1) whichPartDyad=2; // skip spacer and change w2
}
else {
whichPartDyad=(int)(3*runif(0,1));
if (whichPartDyad==3) whichPartDyad--;
}
// if w1 of the dyad (w1-s-w2) is chosen
if (whichPartDyad==0) {
// decide which kmer kmerGroup (4-mer, 5-mer or 6-mer) from which w1 whill be replaced
kmerGroup=(int)(numWordGroup*runif(0,1));
if (kmerGroup==numWordGroup) kmerGroup--;
// select a k-mer from the k-group with prob proportional to z-score
rand=runif(0,1);
tmpDyad[popuCn][0].wordID=0;
for (j=0; j<word[kmerGroup].count; j++) {
if (rand>=word[kmerGroup].prob_sta[j] && rand<=word[kmerGroup].prob_end[j]) {
tmpDyad[popuCn][0].wordID=j; break;
}
}
tmpDyad[popuCn][0].wordGroup=kmerGroup; // mark the kmer kmerGroup from which w1 is selected
}
// if the spacer is chosen
else if (whichPartDyad==1) {
// choose a gap that is different from the current one
do {
spaceWidth=minSpaceWidth+(int)((maxSpaceWidth-minSpaceWidth+1)*runif(0,1));
} while (spaceWidth==tmpDyad[popuCn][1].wordID);
tmpDyad[popuCn][1].wordID=spaceWidth;
tmpDyad[popuCn][1].wordGroup=-1; // dummy - not used
}
// if w2 of the dyad (w1-space-w2) is chosen
else {
kmerGroup=(int)(numWordGroup*runif(0,1));
if (kmerGroup==numWordGroup) kmerGroup--;
rand=runif(0,1);
tmpDyad[popuCn][2].wordID=0;
for (j=0; j<word[kmerGroup].count; j++) {
if (rand>=word[kmerGroup].prob_sta[j] && rand<word[kmerGroup].prob_end[j]) {
tmpDyad[popuCn][2].wordID=j; break;
}
}
tmpDyad[popuCn][2].wordGroup=kmerGroup; // mark the kmer kmerGroup from which w2 is selected
}
}
popuCn++;
}
// update the population
for (i=0; i<populationSize; i++) {
for (j=0; j<3; j++) {
dyad[i][j].wordID =tmpDyad[i][j].wordID;
dyad[i][j].wordGroup=tmpDyad[i][j].wordGroup;
}
maxpFactor[i]=tmpmaxpFactor[i];
}
if (tmpDyad[0]) { free(tmpDyad[0]); tmpDyad[0]=NULL; }
if (tmpDyad) { free(tmpDyad); tmpDyad=NULL; }
if (tmpmaxpFactor) { free(tmpmaxpFactor); tmpmaxpFactor=NULL; }
}
void crossover(Chrs **dyad,int numWordGroup,Words *word,int minSpaceWidth,int maxSpaceWidth,
Wheel *wheel,int populationSize,Fitness *fitness,char *motifUniq,double *maxpFactor,double maxpMutationRate) {
register int i,j;
int found,id1,id2,popuCn;
double *tmpmaxpFactor;
Chrs **tmpDyad;
double rand,du,newFactor;
tmpDyad=alloc_chrs(populationSize,4);
tmpmaxpFactor=alloc_double(populationSize);
// save a copy of all unique top-ranked spaced dyads
popuCn=0;
for (i=0; i<populationSize; i++) {
if (motifUniq[i]=='1') {
for (j=0; j<3; j++) {
tmpDyad[popuCn][j].wordID =dyad[fitness[i].index][j].wordID;
tmpDyad[popuCn][j].wordGroup=dyad[fitness[i].index][j].wordGroup;
}
tmpmaxpFactor[popuCn]=maxpFactor[fitness[i].index];
popuCn++;
}
}
// crossover
while (popuCn<populationSize) {
do {
// select the first dyad
du=(double)populationSize*genrand();
found=0; id1=0; id2=0;
for (j=0; j<populationSize; j++) {
if (du>=wheel[j].start && du<=wheel[j].end) { id1=wheel[j].index; found=1; break; }
}
if (!found) {
id1=(int)(populationSize*genrand());
if (id1==populationSize) id1--;
}
// select the second dyad
du=(double)populationSize*genrand();
found=0;
for (j=0; j<populationSize; j++) {
if (du>=wheel[j].start && du<=wheel[j].end) { id2=wheel[j].index; found=1; break; }
}
if (!found) {
id2=(int)(populationSize*genrand());
if (id2==populationSize) id2--;
}
// printf("id1 id2: %d %d\n",id1,id2);
} while (id1==id2); // do not pick the same "chromosome"
if (genrand()<maxpMutationRate) {
// if change maxp factor other than crossing over the dyads
do {
newFactor=MAXP_BASE+MAXP_FACTOR*((int)(MAXP_SCALE*genrand()));
} while (fabs(newFactor-maxpFactor[id1])<0.001);
tmpmaxpFactor[popuCn]=newFactor;
tmpDyad[popuCn][0].wordID =dyad[id1][0].wordID;
tmpDyad[popuCn][1].wordID =dyad[id1][1].wordID;
tmpDyad[popuCn][2].wordID =dyad[id1][2].wordID;
tmpDyad[popuCn][0].wordGroup=dyad[id1][0].wordGroup;
tmpDyad[popuCn][1].wordGroup=dyad[id1][1].wordGroup;
tmpDyad[popuCn][2].wordGroup=dyad[id1][2].wordGroup;
popuCn++;
if (popuCn==populationSize) break;
do {
newFactor=MAXP_BASE+MAXP_FACTOR*((int)(MAXP_SCALE*genrand()));
} while (fabs(newFactor-maxpFactor[id2])<0.001);
tmpmaxpFactor[popuCn]=newFactor;
tmpDyad[popuCn][0].wordID =dyad[id2][0].wordID;
tmpDyad[popuCn][1].wordID =dyad[id2][1].wordID;
tmpDyad[popuCn][2].wordID =dyad[id2][2].wordID;
tmpDyad[popuCn][0].wordGroup=dyad[id2][0].wordGroup;
tmpDyad[popuCn][1].wordGroup=dyad[id2][1].wordGroup;
tmpDyad[popuCn][2].wordGroup=dyad[id2][2].wordGroup;
popuCn++;
if (popuCn==populationSize) break;
}
else {
rand=genrand();
if (rand>=0 && rand<1/3) {
// replace w1 of dyad1 by w1 of dyad2
tmpDyad[popuCn][0].wordID =dyad[id2][0].wordID; // replacement
tmpDyad[popuCn][1].wordID =dyad[id1][1].wordID;
tmpDyad[popuCn][2].wordID =dyad[id1][2].wordID;
tmpDyad[popuCn][0].wordGroup=dyad[id2][0].wordGroup; // replacement
tmpDyad[popuCn][1].wordGroup=dyad[id1][1].wordGroup;
tmpDyad[popuCn][2].wordGroup=dyad[id1][2].wordGroup;
tmpmaxpFactor[popuCn]=maxpFactor[id1];
popuCn++;
if (popuCn==populationSize) break;
// replace w1 of dyad2 by w1 of dyad1
tmpDyad[popuCn][0].wordID =dyad[id1][0].wordID; // replacement
tmpDyad[popuCn][1].wordID =dyad[id2][1].wordID;
tmpDyad[popuCn][2].wordID =dyad[id2][2].wordID;
tmpDyad[popuCn][0].wordGroup=dyad[id1][0].wordGroup; // replacement
tmpDyad[popuCn][1].wordGroup=dyad[id2][1].wordGroup;
tmpDyad[popuCn][2].wordGroup=dyad[id2][2].wordGroup;
tmpmaxpFactor[popuCn]=maxpFactor[id2];
popuCn++;
if (popuCn==populationSize) break;
}
else if (rand>=1/3 && rand<2/3) {
// replace spacer of dyad1 by spacer of dyad2
tmpDyad[popuCn][0].wordID =dyad[id1][0].wordID;
tmpDyad[popuCn][1].wordID =dyad[id2][1].wordID; // replacement
tmpDyad[popuCn][2].wordID =dyad[id1][2].wordID;
tmpDyad[popuCn][0].wordGroup=dyad[id1][0].wordGroup;
tmpDyad[popuCn][1].wordGroup=dyad[id2][1].wordGroup; // replacement
tmpDyad[popuCn][2].wordGroup=dyad[id1][2].wordGroup;
tmpmaxpFactor[popuCn]=maxpFactor[id1];
popuCn++;
if (popuCn==populationSize) break;
// replace spacer dyad2 by spacer of dyad1
tmpDyad[popuCn][0].wordID =dyad[id2][0].wordID;
tmpDyad[popuCn][1].wordID =dyad[id1][1].wordID; // replacement
tmpDyad[popuCn][2].wordID =dyad[id2][2].wordID;
tmpDyad[popuCn][0].wordGroup=dyad[id2][0].wordGroup;
tmpDyad[popuCn][1].wordGroup=dyad[id1][1].wordGroup; // replacement
tmpDyad[popuCn][2].wordGroup=dyad[id2][2].wordGroup;
tmpmaxpFactor[popuCn]=maxpFactor[id2];
popuCn++;
if (popuCn==populationSize) break;
}
else {
// replace w2 of dyad1 by w2 of dyad2
tmpDyad[popuCn][0].wordID =dyad[id1][0].wordID;
tmpDyad[popuCn][1].wordID =dyad[id1][1].wordID;
tmpDyad[popuCn][2].wordID =dyad[id2][2].wordID; // replacement
tmpDyad[popuCn][0].wordGroup=dyad[id1][0].wordGroup;
tmpDyad[popuCn][1].wordGroup=dyad[id1][1].wordGroup;
tmpDyad[popuCn][2].wordGroup=dyad[id2][2].wordGroup; // replacement
tmpmaxpFactor[popuCn]=maxpFactor[id1];
popuCn++;
if (popuCn==populationSize) break;
// replace w2 of dyad2 by w2 of dyad1
tmpDyad[popuCn][0].wordID =dyad[id2][0].wordID;
tmpDyad[popuCn][1].wordID =dyad[id2][1].wordID;
tmpDyad[popuCn][2].wordID =dyad[id1][2].wordID; // replacement
tmpDyad[popuCn][0].wordGroup=dyad[id2][0].wordGroup;
tmpDyad[popuCn][1].wordGroup=dyad[id2][1].wordGroup;
tmpDyad[popuCn][2].wordGroup=dyad[id1][2].wordGroup; // replacement
tmpmaxpFactor[popuCn]=maxpFactor[id2];
popuCn++;
if (popuCn==populationSize) break;
}
}
// printf("popuCn: %d\n",popuCn);
}
// update the population - both pwm lengths and members
for (i=0; i<populationSize; i++) {
for (j=0; j<3; j++) {
dyad[i][j].wordID =tmpDyad[i][j].wordID;
dyad[i][j].wordGroup=tmpDyad[i][j].wordGroup;
}
maxpFactor[i]=tmpmaxpFactor[i];
}
if (tmpDyad[0]) { free(tmpDyad[0]); tmpDyad[0]=NULL; }
if (tmpDyad) { free(tmpDyad); tmpDyad=NULL; }
if (tmpmaxpFactor) { free(tmpmaxpFactor); tmpmaxpFactor=NULL; }
}
SEXP GADEM_Analysis(SEXP sequence,SEXP sizeSeq, SEXP accession, SEXP Rverbose,SEXP RnumWordGroup,SEXP RnumTop3mer,SEXP RnumTop4mer,SEXP RnumTop5mer,SEXP RnumGeneration,SEXP RpopulationSize, SEXP RpValue,SEXP ReValue,SEXP RextTrim,SEXP RminSpaceWidth,SEXP RmaxSpaceWidth,SEXP RuseChIPscore,SEXP RnumEM,SEXP RfEM, SEXP RwidthWt,SEXP RfullScan, SEXP RslideWinPWM,SEXP RstopCriterion,SEXP RnumBackgSets,SEXP RweightType,SEXP RbFileName,SEXP RListPWM,SEXP RminSites,SEXP RmaskR,SEXP Rnmotifs)
{
char *bFileName;
SEXP ResultsGadem;
SEXP RSpwm;
PROTECT(ResultsGadem=NEW_LIST(100));
int increment=0;
double testrand;
//Number of sequences
int numSeq = INTEGER_VALUE(sizeSeq);
// const
// char *Fastaheader[size];
int incr=0;
int longueur=length(sequence);
int IncrementTemp=0;
// basic settings/info
int maxSeqLen,*seqLen; // sequence info
double aveSeqLen; // sequence info
char **seq,**rseq;
int *geneID; // sequence info
char **oseq,**orseq; // copy of the original sequences
char **sseq,**rsseq; // simulated seqs.
double *bfreq1, *bfreq0=NULL; // base frequencies
double *ChIPScore; // chip score
int maskR; // mask simple repeats before running the algorithm
// pwms
double ***pwm; // initial population of PWMs from spaced dyads
int *pwmLen; // initial pwm lengths
double **opwm2; // EM-derived PWM
double ***opwm; // observed PWMs from identified sites
double ***epwm; // em-optimized PWMs
double **logepwm; // log(em-optimized PWM)
int *pwmnewLen; // final motif length after extending to both ends
// llr score distr.
Pgfs *llrDist; // llr distribution from pgf method
int llrDim; // llr distribution dimension
int **ipwm; // integer pwm for computing llr score distribution
// EM, motif, sites
double pvalueCutoff; // user input, used to determine score cutoff based on ipwm
int *scoreCutoff; // pwm score cutoff for the corresponding p-value cutoff
double logev; // log of E-value of a motif;
int useChIPscore; // indicator for using ChIP-seq score for seq. selection for EM
int numEM; // number of EM steps
double E_valueCutoff; // log E-value cutoff
//int nsitesEM; // number of binding sites in sequences subjected to EM
int minsitesEM; // minimal number of sites in a motif in EM sequences
int *nsites; // number of binding sites in full data
int minsites; // minimal number of sites in a motif in full data
Sites **site; // binding sites in all sequences
int motifCn; // number of motifs sought and found
int extTrim;
int noMotifFound; // none of the dyads in the population resulted in a motif
char **pwmConsensus; // consensus sequences of motifs
double pwmDistCutoff; // test statistic for motif pwm similarity
char *uniqMotif; // motifs in a population unique or not
int numUniq; // number of unique motifs in a population
int slideWinPWM; // sliding window for comparing pwm similarity
int widthWt; // window width in which nucleotides are given large weights for PWM optimization
int fullScan; // scan scan on the original sequences or masked sequences
// background
int numBackgSets;
// weights
double **posWeight; // spatial weights
int weightType; // four weight types 0, 1, 2, 3, or 4
// words for spaced dyad
Words *word; // top-ranked k-mers as the words for spaced dyads
int numTop3mer,numTop4mer,numTop5mer; // No. of top-ranked k-mers as words for dyads
int maxWordSize; // max of the above three
int numWordGroup; // number of non-zero k-mer groups
int minSpaceWidth,maxSpaceWidth; // min and max width of spacer of the spaced dyads
Chrs **dyad; // initial population of "chromosomes"
char **sdyad; // char of spaced dyads
// GA
int populationSize,numGeneration; // GA parameters
double maxpMutationRate;
Fitness *fitness; // "chromosome" fitness
Wheel *wheel; // roulette-wheel selection
// to speed up only select a subset of sequences for EM algorithm
double fEM; // percentage of sequences used in EM algorithm
int numSeqEM; // number of sequences subject to EM
char *Iseq; // Indicator if a sequence is used in EM or not
int *emSeqLen; // length of sequences used in EM
double *maxpFactor;
int numCycle; // number of GADEM cycles
int generationNoMotif; // maximal number of GA generations in a GADEM cycle resulted in no motifs
// mis.
//seed_t seed; // random seed
int motifCn2,id,numCycleNoMotif,verbose,minminSites,nmotifs;
int startPWMfound,stopCriterion;
char *mFileName,*oFileName,*pwmFileName,*tempRbFileName;
time_t start;
int cn[4],bcn[4],*seqCn,*bseqCn,avebnsites,avebnsiteSeq,totalSitesInput;
int i;
int ii=0;
int jjj=0;
/*************/
FILE * output = fopen("output.txt", "w");
/*************/
GetRNGstate();
mFileName=alloc_char(500); mFileName[0]='\0';
oFileName=alloc_char(500); oFileName[0]='\0';
pwmFileName=alloc_char(500); pwmFileName[0]='\0';
bFileName=alloc_char(500); bFileName[0]='\0';
//tempRbFileName=alloc_char(500); tempRbFileName[0]='\0';
seq=NULL; aveSeqLen=0; maxSeqLen=0;
//minsites=-1;
startPWMfound=0;
maxSeqLen=0;
for(incr=1;incr<longueur;incr=incr+2)
{
if (length(STRING_ELT(sequence,(incr)))>maxSeqLen) maxSeqLen=length(STRING_ELT(sequence,(incr)));
}
// fprintf(output,"maxLength=%d",maxSeqLen);
// exit(0);
seq=alloc_char_char(numSeq,maxSeqLen+1);
for(incr=1;incr<longueur;incr=incr+2)
{
for (int j=0; j<length(STRING_ELT(sequence,(incr))); j++)
{
seq[IncrementTemp][j]=CHAR(STRING_ELT(sequence,(incr)))[j];
}
IncrementTemp++;
}
verbose=LOGICAL_VALUE(Rverbose);
numWordGroup=INTEGER_VALUE(RnumWordGroup);
minsites=INTEGER_VALUE(RminSites);
numTop3mer=INTEGER_VALUE(RnumTop3mer);
numTop4mer=INTEGER_VALUE(RnumTop4mer);
numTop5mer=INTEGER_VALUE(RnumTop5mer);
numGeneration=INTEGER_VALUE(RnumGeneration);
populationSize=INTEGER_VALUE(RpopulationSize);
pvalueCutoff=NUMERIC_VALUE(RpValue);
E_valueCutoff=NUMERIC_VALUE(ReValue);
extTrim=INTEGER_VALUE(RextTrim);
minSpaceWidth=INTEGER_VALUE(RminSpaceWidth);
maxSpaceWidth=INTEGER_VALUE(RmaxSpaceWidth);
useChIPscore=NUMERIC_VALUE(RuseChIPscore);
numEM=INTEGER_VALUE(RnumEM);
fEM=NUMERIC_VALUE(RfEM);
widthWt=INTEGER_VALUE(RwidthWt);
fullScan=INTEGER_VALUE(RfullScan);
slideWinPWM=INTEGER_VALUE(RslideWinPWM);
numUniq=populationSize;
stopCriterion=INTEGER_VALUE(RstopCriterion);
numBackgSets=INTEGER_VALUE(RnumBackgSets);
weightType=NUMERIC_VALUE(RweightType);
//const char *tempRbFileName[1];
tempRbFileName = convertRString2Char(RbFileName);
//tempRbFileName[0]=CHAR(STRING_ELT(RbFileName,0));
nmotifs = INTEGER_VALUE(Rnmotifs);
maskR = INTEGER_VALUE(RmaskR);
if(numSeq>MAX_NUM_SEQ)
{
error("Error: maximal number of seqences reached!\nPlease reset MAX_NUM_SEQ in gadem.h and rebuild (see installation)\n");
}
strcpy(bFileName,tempRbFileName);
ChIPScore=alloc_double(MAX_NUM_SEQ);
seqLen=alloc_int(MAX_NUM_SEQ);
geneID=alloc_int(MAX_NUM_SEQ);
// seq=sequences;
// numSeq=size;
int len;
for (i=0; i<numSeq; i++)
{
len=strlen(seq[i]);
seqLen[i]=len;
geneID[i]=INTEGER(accession)[i];
}
aveSeqLen=0;
for (i=0; i<numSeq; i++) aveSeqLen +=seqLen[i]; aveSeqLen /=(double)numSeq;
for (i=0; i<numSeq; i++) {
if (seqLen[i]>maxSeqLen) maxSeqLen=seqLen[i];
}
rseq=alloc_char_char(numSeq,maxSeqLen+1);
oseq=alloc_char_char(numSeq,maxSeqLen+1);
orseq=alloc_char_char(numSeq,maxSeqLen+1);
for (i=0; i<numSeq; i++)
{
if(seqLen[i]>maxSeqLen) maxSeqLen=seqLen[i];
}
reverse_seq(seq,rseq,numSeq,seqLen);
// make a copy of the original sequences both strands
for (i=0; i<numSeq; i++)
{
for (int j=0; j<seqLen[i]; j++)
{
oseq[i][j]=seq[i][j];
orseq[i][j]=rseq[i][j];
}
oseq[i][seqLen[i]]='\0'; orseq[i][seqLen[i]]='\0';
}
if (strcmp(bFileName,"NULL")!= 0)
{
bfreq0=alloc_double(5);
read_background(bFileName,bfreq0);
}
if (GET_LENGTH(RListPWM)!= 0)
{
startPWMfound=1;
}
else { }
// check for input parameters
if(numGeneration<1)
{
error("number of generaton < 1.\n");
}
if(populationSize<1)
{
error("population size < 1.\n");
}
if (minSpaceWidth<0)
{
error("minimal number of unspecified bases in spaced dyads <0.\n");
}
if (maxSpaceWidth<0)
{
error("maximal number of unspecified bases in spaced dyads <0.\n");
}
if (minSpaceWidth>maxSpaceWidth)
{
error("mingap setting must <= to maxgap setting.\n\n");
}
if (maxSpaceWidth+12>MAX_PWM_LENGTH)
{
error("maxgap setting plus word lengths exceed <MAX_PWM_LENGTH>.\n");
}
if (numEM<0)
{
error("number of EM steps is zero.\n");
}
if (numEM==0)
{
error("number of EM steps = 0, no EM optimization is carried out.\n");
}
if (fullScan!=0 && fullScan!=1)
fullScan=0;
maxWordSize=0;
if (numTop3mer>maxWordSize) maxWordSize=numTop3mer;
if (numTop4mer>maxWordSize) maxWordSize=numTop4mer;
if (numTop5mer>maxWordSize) maxWordSize=numTop5mer;
// any one, two or three: tetramer, pentamer, hexamer
if (numTop3mer==0 && numTop4mer==0 && numTop5mer==0)
{
error("maxw3, maxw4, and maxw5 all zero - no words for spaced dyads.\n");
}
// if (startPWMfound && fEM!=0.5 && fEM!=1.0 & verbose)
// {
// warning("fEM argument is ignored in a seeded analysis\n");
// }
if (startPWMfound)
{
// if(verbose)
// {
// if (populationSize!=10 && populationSize!=100) warning("pop argument is ignored in a seeded analysis, -pop is set to 10.\n");
// if (numGeneration!=1 && numGeneration!=5) warning("gen argument is ignored in a seeded analysis, -gen is set to 1.\n");
// }
fEM=1.0;
populationSize=FIXED_POPULATION; numGeneration=1;
}
// number of sequences for EM
if (fEM>1.0 || fEM<=0.0)
{
error("The fraction of sequences subject to EM is %3.2f.\n",fEM);
}
numSeqEM=(int)(fEM*numSeq);
// memory callocations
Iseq =alloc_char(numSeq+1);
opwm2 =alloc_double_double(MAX_PWM_LENGTH,4);
ipwm =alloc_int_int(MAX_PWM_LENGTH,4);
logepwm=alloc_double_double(MAX_PWM_LENGTH,4);
emSeqLen=alloc_int(numSeqEM);
scoreCutoff=alloc_int(1000);
// scoreCutoff=alloc_int(populationSize);
llrDist=alloc_distr(MAX_DIMENSION);
posWeight=alloc_double_double(numSeq,maxSeqLen);
sseq=alloc_char_char(MAX_NUM_SEQ,maxSeqLen+1);
rsseq=alloc_char_char(MAX_NUM_SEQ,maxSeqLen+1);
bfreq1=base_frequency(numSeq,seq,seqLen);
if (strcmp(bFileName,"NULL") == 0)
{
bfreq0=alloc_double(5);
for (i=0; i<4; i++)
{
bfreq0[i]=bfreq1[i];
}
}
// if minN not specified, set the defaults accordingly
if (minsites==-1)
{
minsites =max(2,(int)(numSeq/20));
}
minsitesEM=(int)(fEM*minsites);
maxpMutationRate=MAXP_MUTATION_RATE;
// determine the distribution and critical cut point
pwmDistCutoff=vector_similarity();
/*---------- select a subset of sequences for EM only --------------*/
if (useChIPscore==1)
{
select_high_scoring_seq_for_EM (ChIPScore,numSeq,numSeqEM,Iseq,fEM);
}
else
{
sample_without_replacement(Iseq,numSeqEM,numSeq);
}
/*-------------------- end of selection --------------------------*/
if (maskR==1) mask_repetitive(geneID,seq,numSeq,seqLen,mFileName);
if (widthWt<20)
{
warning("The window width of sequence centered on the nucleotides having large weights in EM for PWM optimization is small\n Motif longer than %d will not be discovered\n",widthWt);
}
time(&start);
// if (weightType==1 || weightType==3)
//ffprintf(output,fp,"window width of sequence centered on the nucleotides having large weights for PWM optimization: %d\n",widthWt);
//ffprintf(output,fp,"pwm score p-value cutoff for declaring binding site:\t%e\n",pvalueCutoff);
if(verbose)
{
ffprintf(output,output,"==============================================================================================\n");
ffprintf(output,output,"input sequence file: %s\n",mFileName);
fprintf(output,"number of sequences and average length:\t\t\t\t%d %5.1f\n",numSeq,aveSeqLen);
fprintf(output,"Use pgf method to approximate llr null distribution\n");
fprintf(output,"parameters estimated from sequences in: %s\n\n",mFileName);
if (weightType!=0)
fprintf(output,"non-uniform weight applies to each sequence - type:\t\t%d\n",weightType);
fprintf(output,"number of GA generations & population size:\t\t\t%d %d\n\n",numGeneration,populationSize);
fprintf(output,"PWM score p-value cutoff for binding site declaration:\t\t%e\n",pvalueCutoff);
fprintf(output,"ln(E-value) cutoff for motif declaration:\t\t\t%f\n\n",E_valueCutoff);
// fprintf(output,"number (percentage) of sequences selected for EM:\t\t%d(%4.1f\%)\n",numSeqEM,100.0*(double)numSeqEM/(double)numSeq);
fprintf(output,"number of EM steps:\t\t\t\t\t\t%d\n",numEM);
fprintf(output,"minimal no. sites considered for a motif:\t\t\t%d\n\n",minsites);
fprintf(output,"[a,c,g,t] frequencies in input data:\t\t\t\t%f %f %f %f\n",bfreq1[0],bfreq1[1],bfreq1[2],bfreq1[3]);
fprintf(output,"==============================================================================================\n");
}
// if (pgf)
// {
// if (userMarkovOrder!=0 & verbose)
// {
// warning("The user-specified background Markov order (%d) is ignored when -pgf is set to 1\n",userMarkovOrder);
// }
// if (bFileName[0]!='\0' & verbose)
// {
// warning("The user-specified background models: %s are not used when -pgf is set to 1\n",bFileName);
// }
// }
// if (startPWMfound && fEM!=1.0 & verbose)
// {
// warning("fEM argument is ignored in a seeded analysis\n");
// }
// determine seq length by counting only [a,c,g,t], seqLen is used in E-value calculation
// determine the distribution and critical cut point
pwmDistCutoff=vector_similarity();
if (weightType==0) assign_weight_uniform(seqLen,numSeq,posWeight);
else if (weightType==1) assign_weight_triangular(seqLen,numSeq,posWeight);
else if (weightType==2) assign_weight_normal(seqLen,numSeq,posWeight);
else
{
error("Motif prior probability type not found - please choose: 0, 1, or 2\n");
// fprintf(output,"Consider: -posWt 1 for strong central enrichment as in ChIP-seq\n");
// fprintf(output," -posWt 0 for others\n\n");
// exit(0);
}
/* if (startPWMfound) minminSites=minsites;
else minminSites=(int)(0.40*minsitesEM);*/
motifCn=0; noMotifFound=0; numCycle=0; numCycleNoMotif=0;
int compt=0;
int lengthList=GET_LENGTH(RListPWM);
/****************************************/
broadcastOnce(maxSeqLen, numEM, startPWMfound, minminSites, maxpFactor, numSeq, numSeqEM, Iseq, bfreq0, posWeight, weightType, pvalueCutoff, emSeqLen, populationSize);
/****************************************/
do
{
if(!startPWMfound)
{
if(verbose)
{
fprintf(output,"*** Running an unseeded analysis ***\n");
// fprintf(output,"\n|------------------------------------------------------------------|\n");
// fprintf(output,"| |\n");
// fprintf(output,"| *** Running an unseeded analysis *** |\n");
// fprintf(output,"| |\n");
// fprintf(output,"|------------------------------------------------------------------|\n\n");
}
populationSize=INTEGER_VALUE(RpopulationSize);
numGeneration=INTEGER_VALUE(RnumGeneration);
dyad =alloc_chrs(populationSize,4);
wheel =alloc_wheel(populationSize);
fitness=alloc_fitness(populationSize);
maxpFactor=alloc_double(populationSize);
uniqMotif=alloc_char(populationSize+1);
opwm =alloc_double_double_double(populationSize,MAX_PWM_LENGTH,4);
epwm=alloc_double_double_double(populationSize,MAX_PWM_LENGTH,4);
pwmConsensus=alloc_char_char(populationSize,MAX_PWM_LENGTH+1);
pwm =alloc_double_double_double(populationSize,MAX_PWM_LENGTH,4);
pwmLen=alloc_int(populationSize);
sdyad =alloc_char_char(populationSize,MAX_PWM_LENGTH+1);
word =alloc_word(numWordGroup,maxWordSize);
minminSites=(int)(0.40*minsitesEM);
// identify top-ranked k-mers (k=3,4,5) for spaced dyads
if(verbose)
fprintf(output,"GADEM cycle %2d: enumerate and count k-mers... ",numCycle+1);
numWordGroup=word_for_dyad(word,seq,rseq,numSeq,seqLen,bfreq1,&numTop3mer,&numTop4mer,&numTop5mer);
if(verbose)
fprintf(output,"Done.\n");
// generating a "population" of spaced dyads
if(verbose)
fprintf(output,"Initializing GA... ");
initialisation(dyad,populationSize,numWordGroup,word,minSpaceWidth,maxSpaceWidth,maxpFactor);
if(verbose)
fprintf(output,"Done.\n");
}
else
{
if(verbose)
{
fprintf(output,"*** Running an seeded analysis ***\n");
// fprintf(output,"\n|------------------------------------------------------------------|\n");
// fprintf(output,"| |\n");
// fprintf(output,"| *** Running a seeded analysis *** |\n");
// fprintf(output,"| |\n");
// fprintf(output,"|------------------------------------------------------------------|\n\n");
}
populationSize=FIXED_POPULATION;
dyad =alloc_chrs(populationSize,4);
pwm=alloc_double_double_double(populationSize,MAX_PWM_LENGTH,4);
pwmLen=alloc_int(populationSize);
maxpFactor=alloc_double(populationSize);
uniqMotif=alloc_char(populationSize+1);
opwm =alloc_double_double_double(populationSize,MAX_PWM_LENGTH,4);
epwm=alloc_double_double_double(populationSize,MAX_PWM_LENGTH,4);
pwmConsensus=alloc_char_char(populationSize,MAX_PWM_LENGTH+1);
sdyad =alloc_char_char(populationSize,MAX_PWM_LENGTH+1);
word =alloc_word(numWordGroup,maxWordSize);
wheel =alloc_wheel(populationSize);
fitness=alloc_fitness(populationSize);
minminSites=minsites;
int lengthMatrix;
lengthMatrix=GET_LENGTH(VECTOR_ELT(RListPWM,compt));
RSpwm=allocMatrix(REALSXP,4,(lengthMatrix/4));
RSpwm=VECTOR_ELT(RListPWM,compt);
pwmLen[0]=read_pwm0(RSpwm,pwm[0],lengthMatrix);
for(i=1; i<populationSize; i++)
{
for (int j=0; j<pwmLen[0]; j++)
{
for (int k=0; k<4; k++)
{
pwm[i][j][k]=pwm[0][j][k];
}
}
pwmLen[i]=pwmLen[0];
}
for (i=0; i<populationSize; i++)
{
maxpFactor[i]=FIXED_MAXPF*(i+1);
standardize_pwm(pwm[i],pwmLen[i]);
consensus_pwm(pwm[i],pwmLen[i],pwmConsensus[i]);
strcpy(sdyad[i],pwmConsensus[i]);
}
}
generationNoMotif=0;
for (jjj=0; jjj<numGeneration; jjj++)
{
// convert spaced dyads to letter probability matrix
if (!startPWMfound)
{
dyad_to_pwm(word,populationSize,dyad,pwm,pwmLen);
}
/*
DO_APPLY(populationCalculation(maxSeqLen, numEM, fitness+ii,
startPWMfound, minminSites, maxpFactor[ii],
numSeq, numSeqEM, seq, rseq, seqLen, Iseq,
bfreq0, posWeight, weightType,
pvalueCutoff, emSeqLen,
pwm[ii], pwmLen[ii], epwm[ii], opwm[ii],
pwmConsensus[ii], scoreCutoff+ii, sdyad[ii], ii),
populationSize, ii);
*/
/* Create the structure to send to all the other slaves */
broadcastEveryCycle(Iseq, pwm, pwmLen, pwmConsensus, scoreCutoff, sdyad, populationSize);
populationCalculation(maxSeqLen, numEM, fitness+ii,
startPWMfound, minminSites, maxpFactor[ii],
numSeq, numSeqEM, seq, rseq, seqLen, Iseq,
bfreq0, posWeight, weightType,
pvalueCutoff, emSeqLen,
pwm[ii], pwmLen[ii], epwm[ii], opwm[ii],
pwmConsensus[ii], scoreCutoff+ii, sdyad[ii], ii);
/* Receive the analyzed data from all the other slaves and compile them */
//getPopCalcResults(...);
// for (i=0; i<5; i++)
// {
// fprintf(output,"fitness.value=%lf\n",fitness[i].value);
// fprintf(output,"fitness.index=%d\n",fitness[i].index);
// fprintf(output,"maxpfactor=%lf\n",maxpFactor[i]);
// fprintf(output,"scoreCutoff=%d\n",scoreCutoff[i]);
// fprintf(output," spacedDyad: %s\n",sdyad[i]);
//
// for (l=0; l<pwmLen[i]; l++)
// {
// for (m=0; m<4; m++)
// {
// fprintf(output,"opwm[%d][%d][%d]=%lf ",i,l,m,opwm[i][l][m]);
// fprintf(output,"epwm[%d][%d][%d]=%lf ",i,l,m,epwm[i][l][m]);
// fprintf(output,"pwm[%d][%d][%d]=%lf ",i,l,m,pwm[i][l][m]);
// }
// fprintf(output,"\n");
// }
// fprintf(output,"\n");
// }
//
// testrand=runif(0,1);
// fprintf(output,"testrand1=%lf\n",testrand);
if (populationSize>1)
{
sort_fitness(fitness,populationSize);
}
// for (i=0; i<5; i++)
// {
// fprintf(output,"fitness.value=%lf\n",fitness[i].value);
// fprintf(output,"fitness.index=%d\n",fitness[i].index);
// }
numUniq=check_pwm_uniqueness_dist(opwm, pwmLen,
populationSize, fitness,
pwmDistCutoff, E_valueCutoff,
uniqMotif, slideWinPWM);
// for (i=0; i<5; i++)
// {
// fprintf(output,"fitness.value=%lf\n",fitness[i].value);
// fprintf(output,"fitness.index=%d\n",fitness[i].index);
// fprintf(output,"maxpfactor=%lf\n",maxpFactor[i]);
// fprintf(output,"scoreCutoff=%d\n",scoreCutoff[i]);
// fprintf(output," spacedDyad: %s\n",sdyad[i]);
//
// for (l=0; l<pwmLen[i]; l++)
// {
// for (m=0; m<4; m++)
// {
// fprintf(output,"opwm[%d][%d][%d]=%lf",i,l,m,opwm[i][l][m]);
// }
// fprintf(output,"\n");
// }
// fprintf(output,"\n");
// }
if(verbose)
{
fprintf(output,"GADEM cycle[%3d] generation[%3d] number of unique motif: %d\n",numCycle+1,jjj+1,numUniq);
for (i=0; i<populationSize; i++)
{
if (uniqMotif[i]=='1')
{
fprintf(output," spacedDyad: %s ",sdyad[fitness[i].index]);
for (int j=strlen(sdyad[fitness[i].index]); j<maxSpaceWidth+10; j++) fprintf(output," ");
fprintf(output,"motifConsensus: %s ",pwmConsensus[fitness[i].index]);
for (int j=strlen(sdyad[fitness[i].index]); j<maxSpaceWidth+10; j++) fprintf(output," ");
fprintf(output," %3.2f fitness: %7.2f\n",maxpFactor[fitness[i].index],fitness[i].value);
}
}
fprintf(output,"\n");
}
if (jjj<numGeneration-1)
{
// fitness based selection with replacement
roulett_wheel_fitness(fitness,populationSize,wheel);
// mutation and crossover operations
if (populationSize>1)
{
testrand=runif(0,1);
if (testrand>=0.5)
{
mutation(dyad,numWordGroup,word,minSpaceWidth,maxSpaceWidth,wheel,populationSize,fitness,uniqMotif,
maxpFactor,maxpMutationRate);
}
else
{
crossover(dyad,numWordGroup,word,minSpaceWidth,maxSpaceWidth,wheel,populationSize,fitness,uniqMotif, maxpFactor,maxpMutationRate);
}
}
else
{
mutation(dyad,numWordGroup,word,minSpaceWidth,maxSpaceWidth,wheel,populationSize,fitness,uniqMotif, maxpFactor,maxpMutationRate);
}
}
}
if((numCycle+1)< lengthList)
{
compt++;
}
else
{
startPWMfound=0;
}
numCycle++;
site=alloc_site_site(numUniq+1,MAX_SITES);
nsites=alloc_int(numUniq+1);
pwmnewLen=alloc_int(numUniq+1); // after base extension and trimming
seqCn=alloc_int(MAX_NUM_SEQ);
bseqCn=alloc_int(MAX_NUM_SEQ);
// final step user-specified background model is used
motifCn2=0; // motifCn per GADEM cycle
for (ii=0; ii<populationSize; ii++)
{
id=fitness[ii].index;
if(uniqMotif[ii]=='0')
{
continue;
}
// approximate the exact llr distribution using Staden's method
// if(verbose)
// {
// fprintf(output,"Approximate the exact pwm llr score distribution using the pgf method.\n");
// }
log_ratio_to_int(epwm[id],ipwm,pwmLen[id],bfreq0);
// compute score distribution of the (int)PWM using Staden's method
llrDim=pwm_score_dist(ipwm,pwmLen[id],llrDist,bfreq0);
//fprintf(output,"Avant ScoreCutoff %d \n",scoreCutoff[id]);
scoreCutoff[id]=determine_cutoff(llrDist,llrDim,pvalueCutoff);
//fprintf(output,"Apres ScoreCutoff %d \n",scoreCutoff[id]);
if(fullScan)
{
nsites[motifCn2]=scan_llr_pgf(llrDist,llrDim,site[motifCn2],numSeq,oseq,orseq,seqLen,ipwm,pwmLen[id],scoreCutoff[id],bfreq0);
}
else
{
nsites[motifCn2]=scan_llr_pgf(llrDist,llrDim,site[motifCn2],numSeq,seq,rseq,seqLen,ipwm,pwmLen[id],scoreCutoff[id],bfreq0);
}
if (nsites[motifCn2]>=max(2,minsites))
{
for (int j=0; j<numSeq; j++) seqCn[j]=0;
for (int j=0; j<nsites[motifCn2]; j++) seqCn[site[motifCn2][j].seq]++;
for (int j=0; j<4; j++) cn[j]=0;
for (int j=0; j<numSeq; j++)
{
if (seqCn[j]==0) cn[0]++;
if (seqCn[j]==1) cn[1]++;
if (seqCn[j]==2) cn[2]++;
if (seqCn[j]>2) cn[3]++;
}
totalSitesInput=nsites[motifCn2];
if (extTrim)
{
if (fullScan)
{
extend_alignment(site[motifCn2],numSeq,oseq,orseq,seqLen,nsites[motifCn2],pwmLen[id],&(pwmnewLen[motifCn2]));
}
else
{
extend_alignment(site[motifCn2],numSeq,seq,rseq,seqLen,nsites[motifCn2],pwmLen[id],&(pwmnewLen[motifCn2]));
}
}
else
{
pwmnewLen[motifCn2]=pwmLen[id];
}
if (fullScan)
{
align_sites_count(site[motifCn2],oseq,orseq,nsites[motifCn2],pwmnewLen[motifCn2],opwm2);
}
else
{
align_sites_count(site[motifCn2],seq,rseq,nsites[motifCn2],pwmnewLen[motifCn2],opwm2);
}
standardize_pwm(opwm2,pwmnewLen[motifCn2]);
logev=E_value(opwm2,nsites[motifCn2],bfreq0,pwmnewLen[motifCn2],numSeq,seqLen);
if (logev<=E_valueCutoff)
{
consensus_pwm(opwm2,pwmnewLen[motifCn2],pwmConsensus[id]);
if (fullScan)
{
SET_VECTOR_ELT(ResultsGadem,increment,print_result_R(site[motifCn2],nsites[motifCn2],numSeq,oseq,orseq,seqLen,logev,opwm2,pwmnewLen[motifCn2],motifCn+1,sdyad[id],pwmConsensus[id],numCycle,pvalueCutoff,maxpFactor[id],geneID));
increment++;
print_motif(site[motifCn2],nsites[motifCn2],oseq,orseq,seqLen,pwmnewLen[motifCn2],motifCn+1,opwm2);
}
else
{
SET_VECTOR_ELT(ResultsGadem,increment,print_result_R(site[motifCn2],nsites[motifCn2],numSeq,seq,rseq,seqLen,logev,opwm2,pwmnewLen[motifCn2],
motifCn+1,sdyad[id],pwmConsensus[id],numCycle,pvalueCutoff,maxpFactor[id],geneID));
increment++;
print_motif(site[motifCn2],nsites[motifCn2],seq,rseq,seqLen,pwmnewLen[motifCn2],motifCn+1,opwm2);
}
mask_sites(nsites[motifCn2],seq,rseq,seqLen,site[motifCn2],pwmnewLen[motifCn2]);
/* ----------------------compute the average number of sites in background sequences ----------------------*/
avebnsites=0; avebnsiteSeq=0;
for (i=0; i<numBackgSets; i++)
{
simulate_background_seq(bfreq0,numSeq,seqLen,sseq);
reverse_seq(sseq,rsseq,numSeq,seqLen);
nsites[motifCn2]=scan_llr_pgf(llrDist,llrDim,site[motifCn2],numSeq,sseq,rsseq,seqLen,ipwm,pwmLen[id],scoreCutoff[id],bfreq0);
for (int j=0; j<numSeq; j++) bseqCn[j]=0;
for (int j=0; j<nsites[motifCn2]; j++) bseqCn[site[motifCn2][j].seq]++;
for (int j=0; j<4; j++) bcn[j]=0;
for (int j=0; j<numSeq; j++)
{
if (bseqCn[j]==0) bcn[0]++;
if (bseqCn[j]==1) bcn[1]++;
if (bseqCn[j]==2) bcn[2]++;
if (bseqCn[j]>2) bcn[3]++;
}
//ffprintf(output,fq,"background set[%2d] Seqs with 0,1,2,>2 sites: %d %d %d %d\n",i+1,bcn[0],bcn[1],bcn[2],bcn[3]);
avebnsites+=nsites[motifCn2]; avebnsiteSeq+=(numSeq-bcn[0]);
}
avebnsites/=numBackgSets; avebnsiteSeq/=numBackgSets;
/* -----------------end compute the average number of sites in background sequences ----------------------*/
motifCn++; motifCn2++;
//if((numCycle+1) > lengthList & fixSeeded)
// {
// numCycleNoMotif=1;
// startPWMfound=1;
// } else {
numCycleNoMotif=0;
// }
}
}
}
/* for (int i=0; i<motifCn2; i++)
{
mask_sites(nsites[i],seq,rseq,seqLen,site[i],pwmnewLen[i]);
} */
if (site[0])
{
free(site[0]);
site[0]=NULL;
}
if (site)
{
free(site);
site=NULL;
}
if (nsites)
{
free(nsites);
nsites=NULL;
}
if (pwmnewLen)
{
free(pwmnewLen);
pwmnewLen=NULL;
}
if (motifCn2==0)
numCycleNoMotif++;
if (motifCn==nmotifs)
{
fprintf(output,"Maximal number of motifs (%d) reached\n",nmotifs);
break;
}
if (numCycleNoMotif==stopCriterion)
noMotifFound=1;
}while (!noMotifFound);
// fclose(fp);
/*if (!startPWMfound) {
if (dyad[0]) { free(dyad[0]); dyad[0]=NULL; }
if (dyad) { free(dyad); dyad=NULL; }
}*/
if (seqLen)
{
free(seqLen);
seqLen=NULL;
}
if (pwm[0][0])
{
free(pwm[0][0]);
pwm[0][0]=NULL;
}
if (pwm[0])
{
free(pwm[0]);
pwm[0]=NULL;
}
if (pwm)
{
free(pwm);
pwm=NULL;
}
if (opwm2[0])
{
free(opwm2[0]);
opwm2[0]=NULL;
}
if (opwm2)
{
free(opwm2);
opwm2=NULL;
}
if (opwm[0][0])
{
free(opwm[0][0]);
opwm[0][0]=NULL;
}
if (opwm[0])
{
free(opwm[0]);
opwm[0]=NULL;
}
if (opwm)
{
free(opwm);
opwm=NULL;
}
if(ipwm[0])
{
free(ipwm[0]);
ipwm[0]=NULL;
}
if (ipwm)
{
free(ipwm);
ipwm=NULL;
}
if (pwmLen)
{
free(pwmLen);
pwmLen=NULL;
}
if (seq[0]) { free(seq[0]); seq[0]=NULL; }
if (seq) { free(seq); seq=NULL; }
// if (rseq[0]) { free(rseq[0]); rseq[0]=NULL; }
// if (rseq) { free(rseq); rseq=NULL; }
// if (oseq[0]) { free(oseq[0]); oseq[0]=NULL; }
// if (oseq) { free(oseq); oseq=NULL; }
// if (orseq[0]) { free(orseq[0]); orseq[0]=NULL; }
// if (orseq) { free(orseq); orseq=NULL; }
if (bfreq1)
{
free(bfreq1);
bfreq1=NULL;
}
if (bfreq0)
{
free(bfreq0);
bfreq0=NULL;
}
if (wheel)
{
free(wheel);
wheel=NULL;
}
if (fitness)
{
free(fitness);
fitness=NULL;
}
if (mFileName)
{
free(mFileName);
mFileName=NULL;
}
if (oFileName)
{
free(oFileName);
oFileName=NULL;
}
if (pwmFileName)
{
free(pwmFileName);
pwmFileName=NULL;
}
if (sdyad[0])
{
free(sdyad[0]);
sdyad[0]=NULL;
}
if (sdyad)
{
free(sdyad);
sdyad=NULL;
}
if (pwmConsensus[0])
{
free(pwmConsensus[0]);
pwmConsensus[0]=NULL;
}
if (pwmConsensus)
{
free(pwmConsensus);
pwmConsensus=NULL;
}
//if (!startPWMfound && word) destroy_word(word,numWordGroup);
PutRNGstate();
UNPROTECT(1);
return(ResultsGadem);
}
void select_high_scoring_seq_for_EM (double *ChIPScore,int numSeq,int numSeqEM,char *Iseq,double fEM) {
register int i;
int numSeqWithQualityScore,numSeqEMtmp1,numSeqEMtmp2;
double *tmpScore;
double ChIPscoreCutoff;
tmpScore=alloc_double(numSeq);
numSeqWithQualityScore=0;
for (i=0; i<numSeq; i++)
{
if (ChIPScore[i]>0) numSeqWithQualityScore++;
}
tmpScore=alloc_double(numSeq);
for (i=0; i<numSeq; i++) tmpScore[i]=ChIPScore[i];
sort_double(tmpScore,numSeq);
ChIPscoreCutoff=tmpScore[(int)(fEM*numSeq)];
if (numSeqWithQualityScore<=(int)(fEM*numSeq))
{
for (i=0; i<numSeq; i++) Iseq[i]='0';
numSeqEMtmp1=0;
for (i=0; i<numSeq; i++)
{
if (ChIPScore[i]>0)
{
Iseq[i]='1'; numSeqEMtmp1++;
}
}
numSeqEMtmp2=0;
for (i=0; i<numSeq; i++)
{
if (ChIPScore[i]<=0)
{
Iseq[i]='1'; numSeqEMtmp2++;
if (numSeqEMtmp1+numSeqEMtmp2==numSeqEM) break;
}
}
}
else
{
for (i=0; i<numSeq; i++) Iseq[i]='0';
numSeqEMtmp1=0; numSeqEMtmp2=0;
for (i=0; i<numSeq; i++) {
if (ChIPScore[i]>=ChIPscoreCutoff)
{
Iseq[i]='1'; numSeqEMtmp1++;
if (numSeqEMtmp1==numSeqEM) break;
}
}
}
if (tmpScore)
{
free(tmpScore);
tmpScore=NULL;
}
if (ChIPScore)
{
free(ChIPScore);
ChIPScore=NULL;
}
}
int evresp_(char *sta, char *cha, char *net, char *locid, char *datime,
char *units, char *file, float *freqs, int *nfreqs_in, float *resp,
char *rtype, char *verbose, int *start_stage, int *stop_stage,
int *stdio_flag, int lsta, int lcha, int lnet, int llocid,
int ldatime, int lunits, int lfile, int lrtype, int lverbose, int useTotalSensitivityFlag)
{
struct response *first = (struct response *)NULL;
double *dfreqs;
int i,j, nfreqs, start, stop, flag;
/* add null characters to end of input string arguments (remove trailing
spaces first */
add_null(sta, lsta-1, 'a');
add_null(cha, lcha-1, 'a');
add_null(net, lnet-1, 'a');
add_null(locid, llocid-1, 'a');
add_null(datime, ldatime-1, 'a');
add_null(units, lunits-1, 'a');
add_null(file, lfile-1, 'a');
add_null(rtype, lrtype-1, 'a');
add_null(verbose, lverbose-1, 'a');
nfreqs = *nfreqs_in;
start = *start_stage;
stop = *stop_stage;
flag = *stdio_flag;
dfreqs = alloc_double(nfreqs);
for(i = 0; i < nfreqs; i++)
dfreqs[i] = freqs[i];
/* then call evresp */
first = evresp(sta, cha, net, locid, datime, units, file, dfreqs, nfreqs,
rtype, verbose, start, stop, flag, useTotalSensitivityFlag);
/* free up the frequency vector */
free(dfreqs);
/* check the output. If no response found, return 1, else if more than one response
found, return -1 */
if(first == (struct response *)NULL) {
return(1);
}
else if(first->next != (struct response *)NULL) {
free_response(first);
return(-1);
}
/* if only one response found, convert from complex output vector into multiplexed
real output for FORTRAN (real1, imag1, real2, imag2, ..., realN, imagN) */
for(i = 0, j = 0; i < nfreqs; i++) {
resp[j++] = (float) first->rvec[i].real;
resp[j++] = (float) first->rvec[i].imag;
}
/* free up dynamically allocated space */
free_response(first);
/* and return to FORTRAN program */
return(0);
}
// Find prototypes by the MST approach
static void
mst_prototypes (struct opf_graph * sg)
{
int p, q;
double weight;
struct real_heap *Q = NULL;
double *path_val = NULL;
int pred;
int nproto;
// initialization
path_val = alloc_double (sg->node_n);
Q = real_heap_create (sg->node_n, path_val);
for (p = 0; p < sg->node_n; p++)
{
path_val[p] = DBL_MAX;
sg->node[p].status = STATUS_NOTHING;
}
path_val[0] = 0;
sg->node[0].pred = NIL;
real_heap_insert (Q, 0);
nproto = 0;
// Prim's algorithm for Minimum Spanning Tree
while (!real_heap_is_empty (Q))
{
real_heap_remove (Q, &p);
assert (p >= 0 && p < sg->node_n);
sg->node[p].path_val = path_val[p];
pred = sg->node[p].pred;
if (pred != NIL)
if (sg->node[p].label_true != sg->node[pred].label_true)
{
if (sg->node[p].status != STATUS_PROTOTYPE)
{
sg->node[p].status = STATUS_PROTOTYPE;
nproto++;
}
if (sg->node[pred].status != STATUS_PROTOTYPE)
{
sg->node[pred].status = STATUS_PROTOTYPE;
nproto++;
}
}
for (q = 0; q < sg->node_n; q++)
{
if (Q->color[q] != COLOR_BLACK)
{
if (p != q)
{
weight = opf_graph_get_distance (sg, &sg->node[p], &sg->node[q]);
if (weight < path_val[q])
{
sg->node[q].pred = p;
real_heap_update (Q, q, weight);
}
}
}
}
}
real_heap_destroy (&Q);
free (path_val);
/* the algorithm will work even if there
is just one class in the training set */
if (nproto == 0)
sg->node[0].status = STATUS_PROTOTYPE;
}
void
opf_supervised_train (struct opf_graph * sg)
{
int p, q, i;
double tmp, weight;
struct real_heap *Q = NULL;
double *path_val = NULL;
// compute optimum prototypes
mst_prototypes (sg);
// initialization
path_val = alloc_double (sg->node_n);
Q = real_heap_create (sg->node_n, path_val);
for (p = 0; p < sg->node_n; p++)
{
if (sg->node[p].status == STATUS_PROTOTYPE)
{
sg->node[p].pred = NIL;
path_val[p] = 0;
sg->node[p].label = sg->node[p].label_true;
real_heap_insert (Q, p);
}
else // non-prototypes
{
path_val[p] = DBL_MAX;
}
}
// IFT with fmax
i = 0;
while (!real_heap_is_empty (Q))
{
real_heap_remove (Q, &p);
assert (p >= 0 && p < sg->node_n);
sg->ordered_list_of_nodes[i] = p;
i++;
sg->node[p].path_val = path_val[p];
for (q = 0; q < sg->node_n; q++)
{
if (p != q)
{
if (path_val[p] < path_val[q])
{
weight = opf_graph_get_distance (sg, &sg->node[p], &sg->node[q]);
tmp = MAX (path_val[p], weight);
if (tmp < path_val[q])
{
sg->node[q].pred = p;
sg->node[q].label = sg->node[p].label;
real_heap_update (Q, q, tmp);
}
}
}
}
}
real_heap_destroy (&Q);
free (path_val);
}