struct hash *loadAndWeighTce(char *fileName, struct hash *refToPepHash,
struct hash *pepToRefHash)
/* Load transcript cds evidence from file into hash of
* txCdsInfo. */
{
/* Read all tce's in file into list and hash of txCdsInfo. */
struct txCdsInfo *tx, *txList = NULL;
struct lineFile *lf = lineFileOpen(fileName, TRUE);
struct hash *hash = hashNew(18);
char *row[CDSEVIDENCE_NUM_COLS];
while (lineFileRow(lf, row))
{
/* Convert row to cdsEvidence structure. */
struct cdsEvidence *cds = cdsEvidenceLoad(row);
char *acc = txAccFromTempName(cds->name);
tx = hashFindVal(hash, cds->name);
if (tx == NULL)
{
AllocVar(tx);
hashAddSaveName(hash, cds->name, tx, &tx->name);
slAddHead(&txList, tx);
}
/* Track whether it's refSeq, and the associated protein. */
char *refSeqAcc = NULL, *refPepAcc = NULL;
refPepAcc = hashFindVal(refToPepHash, acc);
refSeqAcc = hashFindVal(pepToRefHash, acc);
if (refPepAcc != NULL && refSeqAcc == NULL)
refSeqAcc = acc;
if (refSeqAcc != NULL && refPepAcc == NULL)
refPepAcc = acc;
/* If we are refSeq, bump our score for matches to our own
* bits by a huge factor. */
if (refPepAcc != NULL && startsWith("RefPep", cds->source)
&& sameString(cds->accession, refPepAcc))
cds->score += refSeqWeight;
if (refSeqAcc != NULL && startsWith("RefSeq", cds->source)
&& sameString(cds->accession, refSeqAcc))
cds->score += refSeqWeight + 100;
/* If we are CCDS then that's great too. */
if (sameString("ccds", cds->source))
cds->score += ccdsWeight;
/* If we are txCdsPredict bump weight too. Only RefSeq and txCdsPredict
* can actually possibly make it over real threshold. */
if (sameString("txCdsPredict", cds->source))
cds->score += txCdsPredictWeight;
slAddHead(&tx->cdsList, cds);
}
lineFileClose(&lf);
slReverse(&txList);
/* Sort all cdsLists by score. */
for (tx = txList; tx != NULL; tx = tx->next)
slSort(&tx->cdsList, cdsEvidenceCmpScore);
return hash;
}
void txInfoAssemble(char *txBedFile, char *cdsEvFile, char *txCdsPredictFile, char *altSpliceFile,
char *exceptionFile, char *sizePolyAFile, char *pslFile, char *flipFile, char *outFile)
/* txInfoAssemble - Assemble information from various sources into txInfo table.. */
{
/* Build up hash of evidence keyed by transcript name. */
struct hash *cdsEvHash = hashNew(18);
struct cdsEvidence *cdsEv, *cdsEvList = cdsEvidenceLoadAll(cdsEvFile);
for (cdsEv = cdsEvList; cdsEv != NULL; cdsEv = cdsEv->next)
hashAddUnique(cdsEvHash, cdsEv->name, cdsEv);
verbose(2, "Loaded %d elements from %s\n", cdsEvHash->elCount, cdsEvFile);
/* Build up hash of bestorf structures keyed by transcript name */
struct hash *predictHash = hashNew(18);
struct cdsEvidence *predict, *predictList = cdsEvidenceLoadAll(txCdsPredictFile);
for (predict = predictList; predict != NULL; predict = predict->next)
hashAddUnique(predictHash, predict->name, predict);
verbose(2, "Loaded %d predicts from %s\n", predictHash->elCount, txCdsPredictFile);
/* Build up structure for random access of retained introns */
struct bed *altSpliceList = bedLoadNAll(altSpliceFile, 6);
verbose(2, "Loaded %d alts from %s\n", slCount(altSpliceList), altSpliceFile);
struct hash *altSpliceHash = bedsIntoHashOfKeepers(altSpliceList);
/* Read in exception info. */
struct hash *selenocysteineHash, *altStartHash;
genbankExceptionsHash(exceptionFile, &selenocysteineHash, &altStartHash);
/* Read in polyA sizes */
struct hash *sizePolyAHash = hashNameIntFile(sizePolyAFile);
verbose(2, "Loaded %d from %s\n", sizePolyAHash->elCount, sizePolyAFile);
/* Read in psls */
struct hash *pslHash = hashNew(20);
struct psl *psl, *pslList = pslLoadAll(pslFile);
for (psl = pslList; psl != NULL; psl = psl->next)
hashAdd(pslHash, psl->qName, psl);
verbose(2, "Loaded %d from %s\n", pslHash->elCount, pslFile);
/* Read in accessions that we flipped for better splice sites. */
struct hash *flipHash = hashWordsInFile(flipFile, 0);
/* Open primary gene input and output. */
struct lineFile *lf = lineFileOpen(txBedFile, TRUE);
FILE *f = mustOpen(outFile, "w");
/* Main loop - process each gene */
char *row[12];
while (lineFileRow(lf, row))
{
struct bed *bed = bedLoad12(row);
verbose(3, "Processing %s\n", bed->name);
/* Initialize info to zero */
struct txInfo info;
ZeroVar(&info);
/* Figure out name, sourceAcc, and isRefSeq from bed->name */
info.name = bed->name;
info.category = "n/a";
if (isRfam(bed->name) || stringIn("tRNA", bed->name) != NULL)
{
info.sourceAcc = cloneString(bed->name);
}
else
{
info.sourceAcc = txAccFromTempName(bed->name);
}
info.isRefSeq = startsWith("NM_", info.sourceAcc);
if (startsWith("antibody.", info.sourceAcc)
|| startsWith("CCDS", info.sourceAcc) || isRfam(info.sourceAcc)
|| stringIn("tRNA", info.sourceAcc) != NULL)
{
/* Fake up some things for antibody frag and CCDS that don't have alignments. */
info.sourceSize = bedTotalBlockSize(bed);
info.aliCoverage = 1.0;
info.aliIdRatio = 1.0;
info. genoMapCount = 1;
}
else
{
/* Loop through all psl's associated with our RNA. Figure out
* our overlap with each, and pick best one. */
struct hashEl *hel, *firstPslHel = hashLookup(pslHash, info.sourceAcc);
if (firstPslHel == NULL)
errAbort("%s is not in %s", info.sourceAcc, pslFile);
int mapCount = 0;
struct psl *psl, *bestPsl = NULL;
int coverage, bestCoverage = 0;
boolean isFlipped = (hashLookup(flipHash, info.sourceAcc) != NULL);
for (hel = firstPslHel; hel != NULL; hel = hashLookupNext(hel))
{
psl = hel->val;
mapCount += 1;
coverage = pslBedOverlap(psl, bed);
if (coverage > bestCoverage)
{
bestCoverage = coverage;
bestPsl = psl;
}
/* If we flipped it, try it on the opposite strand too. */
if (isFlipped)
{
psl->strand[0] = (psl->strand[0] == '+' ? '-' : '+');
coverage = pslBedOverlap(psl, bed);
if (coverage > bestCoverage)
{
bestCoverage = coverage;
bestPsl = psl;
}
psl->strand[0] = (psl->strand[0] == '+' ? '-' : '+');
}
}
if (bestPsl == NULL)
errAbort("%s has no overlapping alignments with %s in %s",
bed->name, info.sourceAcc, pslFile);
/* Figure out and save alignment statistics. */
int polyA = hashIntValDefault(sizePolyAHash, bed->name, 0);
info.sourceSize = bestPsl->qSize - polyA;
info.aliCoverage = (double)bestCoverage / info.sourceSize;
info.aliIdRatio = (double)(bestPsl->match + bestPsl->repMatch)/
(bestPsl->match + bestPsl->misMatch + bestPsl->repMatch);
info. genoMapCount = mapCount;
}
/* Get orf size and start/end complete from cdsEv. */
if (bed->thickStart < bed->thickEnd)
{
cdsEv = hashFindVal(cdsEvHash, bed->name);
if (cdsEv != NULL)
{
info.orfSize = cdsEv->end - cdsEv->start;
info.startComplete = cdsEv->startComplete;
info.endComplete = cdsEv->endComplete;
}
}
/* Get score from prediction. */
predict = hashFindVal(predictHash, bed->name);
if (predict != NULL)
info.cdsScore = predict->score;
/* Figure out nonsense-mediated-decay from bed itself. */
info.nonsenseMediatedDecay = isNonsenseMediatedDecayTarget(bed);
/* Figure out if retained intron from bed and alt-splice keeper hash */
info.retainedIntron = hasRetainedIntron(bed, altSpliceHash);
info.strangeSplice = countStrangeSplices(bed, altSpliceHash);
info.atacIntrons = countAtacIntrons(bed, altSpliceHash);
info.bleedIntoIntron = addIntronBleed(bed, altSpliceHash);
/* Look up selenocysteine info. */
info.selenocysteine = (hashLookup(selenocysteineHash, bed->name) != NULL);
/* Loop through bed looking for small gaps indicative of frame shift/stop */
int i, lastBlock = bed->blockCount-1;
int exonCount = 1;
for (i=0; i < lastBlock; ++i)
{
int gapStart = bed->chromStarts[i] + bed->blockSizes[i];
int gapEnd = bed->chromStarts[i+1];
int gapSize = gapEnd - gapStart;
switch (gapSize)
{
case 1:
case 2:
info.genomicFrameShift = TRUE;
break;
case 3:
info.genomicStop = TRUE;
break;
default:
exonCount += 1;
break;
}
}
info.exonCount = exonCount;
/* Write info, free bed. */
txInfoTabOut(&info, f);
bedFree(&bed);
}
/* Clean up and go home. */
carefulClose(&f);
}
void txGeneCdsMap(char *inBed, char *inInfo, char *inPicks, char *refPepToTxPsl,
char *refToPepTab, char *chromSizes, char *cdsToRna, char *rnaToGenome)
/* txGeneCdsMap - Create mapping between CDS region of gene and genome. */
{
/* Load info into hash. */
struct hash *infoHash = hashNew(18);
struct txInfo *info, *infoList = txInfoLoadAll(inInfo);
for (info = infoList; info != NULL; info = info->next)
hashAdd(infoHash, info->name, info);
/* Load picks into hash. We don't use cdsPicksLoadAll because empty fields
* cause that autoSql-generated routine problems. */
struct hash *pickHash = newHash(18);
struct cdsPick *pick;
struct lineFile *lf = lineFileOpen(inPicks, TRUE);
char *row[CDSPICK_NUM_COLS];
while (lineFileRowTab(lf, row))
{
pick = cdsPickLoad(row);
hashAdd(pickHash, pick->name, pick);
}
lineFileClose(&lf);
/* Load refPep/tx alignments into hash keyed by tx. */
struct hash *refPslHash = hashNew(18);
struct psl *psl, *pslList = pslLoadAll(refPepToTxPsl);
for (psl = pslList; psl != NULL; psl = psl->next)
hashAdd(refPslHash, psl->tName, psl);
struct hash *refToPepHash = hashTwoColumnFile(refToPepTab);
struct hash *chromSizeHash = hashNameIntFile(chromSizes);
/* Load in bed. */
struct bed *bed, *bedList = bedLoadNAll(inBed, 12);
/* Open output, and stream through bedList, writing output. */
FILE *fCdsToRna = mustOpen(cdsToRna, "w");
FILE *fRnaToGenome = mustOpen(rnaToGenome, "w");
int refTotal = 0, refFound = 0;
for (bed = bedList; bed != NULL; bed = bed->next)
{
if (bed->thickStart < bed->thickEnd)
{
char *chrom = bed->chrom;
int chromSize = hashIntVal(chromSizeHash, chrom);
info = hashMustFindVal(infoHash, bed->name);
pick = hashMustFindVal(pickHash, bed->name);
if (info->isRefSeq)
{
char *refAcc = txAccFromTempName(bed->name);
if (!startsWith("NM_", refAcc))
errAbort("Don't think I did find that refSeq acc, got %s", refAcc);
char *protAcc = hashMustFindVal(refToPepHash, refAcc);
++refTotal;
if (findAndMapPsl(bed, protAcc, refPslHash, chromSize, fCdsToRna))
++refFound;
}
else
{
fakeCdsToMrna(bed, fCdsToRna);
}
fakeRnaToGenome(bed, chromSize, fRnaToGenome);
}
}
verbose(1, "Missed %d of %d refSeq protein mappings. A small number of RefSeqs just map\n"
"to genome in the UTR.\n", refTotal - refFound, refTotal);
carefulClose(&fCdsToRna);
carefulClose(&fRnaToGenome);
}