struct dnaMotif *dnaMotifLoadNamed(struct sqlConnection *conn,
char *table, char *name)
/* Load motif of given name from table. Return NULL if no such
* motif. */
{
char where[256];
safef(where, sizeof(where), "name='%s'", name);
return dnaMotifLoadWhere(conn, table, where);
}
struct dnaMotif *loadDnaMotif(char *motifName, char *motifTable)
/* Load dnaMotif from table. */
{
struct sqlConnection *conn = hAllocConn(database);
char query[256];
struct dnaMotif *motif;
sqlSafefFrag(query, sizeof query, "name = '%s'", motifName);
motif = dnaMotifLoadWhere(conn, motifTable, query);
hFreeConn(&conn);
return motif;
}
void motifFinder(char *database, char *name, int fileCount, char *files[])
/* motifFinder - find largest scoring motif in bed items. */
{
struct sqlConnection *conn = sqlConnect(database);
int fileNum;
char where[256];
struct chromInfo *ci = createChromInfoList(NULL, database);
sqlSafefFrag(where, sizeof(where), "name = '%s'", name);
struct dnaMotif *motif = dnaMotifLoadWhere(conn, motifTable, where);
if(markovTable != NULL)
dnaMotifMakeLog2(motif);
if(motif == NULL)
errAbort("couldn't find motif '%s'", name);
for (fileNum = 0; fileNum < fileCount; fileNum++)
{
char *words[64], *line;
char **row;
struct lineFile *lf = lineFileOpen(files[fileNum], TRUE);
while (lineFileNextReal(lf, &line))
{
int dnaLength, i, j, rowOffset, length, wordCount = chopTabs(line, words);
unsigned chromSize;
boolean markovFound = FALSE;
double mark0[5];
double mark2[5][5][5];
struct dnaSeq *seq = NULL;
char *dupe = NULL;
if (0 == wordCount)
continue;
lineFileExpectAtLeast(lf, 3, wordCount);
dupe = cloneString(line);
char *chrom = words[0];
int chromStart = lineFileNeedNum(lf, words, 1);
if(markovTable != NULL)
chromStart = max(2, chromStart);
unsigned chromEnd = lineFileNeedNum(lf, words, 2);
if (chromEnd < 1)
errAbort("ERROR: line %d:'%s'\nchromEnd is less than 1\n",
lf->lineIx, dupe);
if (chromStart > chromEnd)
errAbort("ERROR: line %d:'%s'\nchromStart after chromEnd (%d > %d)\n",
lf->lineIx, dupe, chromStart, chromEnd);
length = chromEnd - chromStart;
chromSize = getChromSize(ci, chrom);
if(markovTable == NULL)
{
dnaLength = length;
seq = hDnaFromSeq(database, chrom, chromStart, chromEnd, dnaUpper);
if(uniformBackground)
{
int i;
mark0[0] = 1;
for(i = 1; i <= 4; i++)
mark0[i] = 0.25;
}
else
{
dnaMark0(seq, mark0, NULL);
}
}
else
{
dnaLength = length + 4;
if(chromStart - 2 + dnaLength > chromSize)
// can't do analysis for potential peak hanging off the end of the chrom
continue;
seq = hDnaFromSeq(database, chrom, chromStart - 2, chromEnd + 2, dnaUpper);
struct sqlResult *sr = hRangeQuery(conn, markovTable, chrom, chromStart,
chromStart + 1, NULL, &rowOffset);
if((row = sqlNextRow(sr)) != NULL)
{
dnaMark2Deserialize(row[rowOffset + 3], mark2);
dnaMarkMakeLog2(mark2);
markovFound = TRUE;
}
else
errAbort("markov table '%s' is missing; non-markov analysis is current not supported", markovTable);
sqlFreeResult(&sr);
}
struct bed6FloatScore *hits = NULL;
for (i = 0; i < 2; i++)
{
double mark0Copy[5];
char strand = i == 0 ? '+' : '-';
for (j = 0; j <= 4; j++)
mark0Copy[j] = mark0[j];
if(strand == '-')
{
// reverse markov table too!
double tmp;
reverseComplement(seq->dna, dnaLength);
tmp = mark0Copy[1];
mark0Copy[1] = mark0Copy[3];
mark0Copy[3] = tmp;
tmp = mark0Copy[2];
mark0Copy[2] = mark0Copy[4];
mark0Copy[4] = tmp;
}
for (j = 0; j < length - motif->columnCount + 1; j++)
// tricky b/c if(markovFound) then seq includes the two bytes on either side of actual sequence.
{
double score;
if(markovFound)
score = dnaMotifBitScoreWithMarkovBg(motif, seq->dna + j, mark2);
else
score = dnaMotifBitScoreWithMark0Bg(motif, seq->dna + j, mark0Copy);
if(score >= minScoreCutoff)
{
int start;
if(strand == '-')
start = (chromEnd - j) - motif->columnCount;
else
start = chromStart + j;
struct bed6FloatScore *hit = NULL;
// Watch out for overlapping hits (on either strand; yes, I've seen that happen);
// we report only the highest scoring hit in this case.
// O(n^2) where n == number of motifs in a peak, but I expect n to be almost always very small.
if(!originalCoordinates)
{
for (hit = hits; hit != NULL; hit = hit->next)
{
if(hit->chromEnd > start && hit->chromStart <= (start + motif->columnCount))
{
verbose(3, "found overlapping hits: %d-%d overlaps with %d-%d\n", start, start + motif->columnCount, hit->chromStart, hit->chromEnd);
break;
}
}
}
if(hit == NULL || hit->score < score)
{
if(hit == NULL)
{
AllocVar(hit);
slAddHead(&hits, hit);
hit->chrom = cloneString(chrom);
}
hit->chromStart = originalCoordinates ? chromStart : start;
hit->chromEnd = originalCoordinates ? chromEnd : start + motif->columnCount;
hit->score = score;
hit->strand[0] = strand;
}
}
verbose(3, "j: %d; score: %.2f\n", j, score);
}
}
slSort(&hits, bed6FloatCmpDesc);
int count;
float currentPrior = prior;
for(count = 1; hits != NULL; count++, hits = hits->next)
{
if(topOnly && count > topOnly)
break;
// Use a progressively weaker prior for hits with lower scores
verbose(3, "count: %d; score: %.2f; prior: %.2f; log2(prior / (1 - prior)): %.2f\n", count, hits->score, currentPrior, log2(currentPrior / (1 - currentPrior)));
if(hits->score >= minScoreCutoff - log2(currentPrior / (1 - currentPrior)))
{
printf("%s\t%d\t%d\t%s\t%.2f\t%c\n", chrom, originalCoordinates ? chromStart : hits->chromStart,
originalCoordinates ? chromEnd : hits->chromStart + motif->columnCount, name, hits->score, hits->strand[0]);
currentPrior = count == 1 ? priorBackoff : currentPrior * priorBackoff;
if(count > 2)
verbose(3, "hit for count: %d at %s:%d-%d\n", count, chrom, hits->chromStart, hits->chromStart + motif->columnCount);
}
else
break;
}
freeDnaSeq(&seq);
freeMem(dupe);
}
lineFileClose(&lf);
}
sqlDisconnect(&conn);
}
