/**
* Compare two records based on their duplicate scores. If the scores are equal, we break ties
* based on mapping quality (added to the mate's mapping quality if paired and mapped), then
* library/read name.
*
* <p>If true is given to assumeMateCigar, then any score that can use the mate cigar to to
* compute the mate's score will return the score computed on both ends.
*
* <p>We allow different scoring strategies. We return <0 if rec1 has a better strategy than rec2.
*/
public static int compare(
final SAMRecord rec1,
final SAMRecord rec2,
final ScoringStrategy scoringStrategy,
final boolean assumeMateCigar) {
int cmp;
// always prefer paired over non-paired
if (rec1.getReadPairedFlag() != rec2.getReadPairedFlag())
return rec1.getReadPairedFlag() ? 1 : -1;
cmp =
computeDuplicateScore(rec2, scoringStrategy, assumeMateCigar)
- computeDuplicateScore(rec1, scoringStrategy, assumeMateCigar);
/**
* Finally, use library ID and read name This is important because we cannot control the order
* in which reads appear for reads that are comparable up to now (i.e. cmp == 0). We want to
* deterministically choose them, and so we need this.
*/
if (0 == cmp)
cmp = SAMUtils.getCanonicalRecordName(rec1).compareTo(SAMUtils.getCanonicalRecordName(rec2));
return cmp;
}
/**
* Returns the duplicate score computed from the given fragment. value should be capped by
* Short.MAX_VALUE/2 since the score from two reads will be added and an overflow will be
*
* <p>If true is given to assumeMateCigar, then any score that can use the mate cigar to compute
* the mate's score will return the score computed on both ends.
*/
public static short computeDuplicateScore(
final SAMRecord record,
final ScoringStrategy scoringStrategy,
final boolean assumeMateCigar) {
Short storedScore = (Short) record.getTransientAttribute(Attr.DuplicateScore);
if (storedScore == null) {
short score = 0;
switch (scoringStrategy) {
case SUM_OF_BASE_QUALITIES:
// two (very) long reads worth of high-quality bases can go over Short.MAX_VALUE/2
// and risk overflow.
score += (short) Math.min(getSumOfBaseQualities(record), Short.MAX_VALUE / 2);
break;
case TOTAL_MAPPED_REFERENCE_LENGTH:
if (!record.getReadUnmappedFlag()) {
// no need to remember the score since this scoring mechanism is symmetric
score = (short) Math.min(record.getCigar().getReferenceLength(), Short.MAX_VALUE / 2);
}
if (assumeMateCigar && record.getReadPairedFlag() && !record.getMateUnmappedFlag()) {
score +=
(short)
Math.min(
SAMUtils.getMateCigar(record).getReferenceLength(), Short.MAX_VALUE / 2);
}
break;
// The RANDOM score gives the same score to both reads so that they get filtered together.
// it's not critical do use the readName since the scores from both ends get added, but it
// seem
// to be clearer this way.
case RANDOM:
// start with a random number between Short.MIN_VALUE/4 and Short.MAX_VALUE/4
score += (short) (hasher.hashUnencodedChars(record.getReadName()) & 0b11_1111_1111_1111);
// subtract Short.MIN_VALUE/4 from it to end up with a number between
// 0 and Short.MAX_VALUE/2. This number can be then discounted in case the read is
// not passing filters. We need to stay far from overflow so that when we add the two
// scores from the two read mates we do not overflow since that could cause us to chose a
// failing read-pair instead of a passing one.
score -= Short.MIN_VALUE / 4;
}
// make sure that filter-failing records are heavily discounted. (the discount can happen
// twice, once
// for each mate, so need to make sure we do not subtract more than Short.MIN_VALUE overall.)
score += record.getReadFailsVendorQualityCheckFlag() ? (short) (Short.MIN_VALUE / 2) : 0;
storedScore = score;
record.setTransientAttribute(Attr.DuplicateScore, storedScore);
}
return storedScore;
}
public void acceptRecord(final SAMRecordAndReference args) {
final SAMRecord rec = args.getSamRecord();
final ReferenceSequence ref = args.getReferenceSequence();
if (rec.getReadPairedFlag()) {
if (rec.getFirstOfPairFlag()) {
firstOfPairCollector.addRecord(rec, ref);
} else {
secondOfPairCollector.addRecord(rec, ref);
}
pairCollector.addRecord(rec, ref);
} else {
unpairedCollector.addRecord(rec, ref);
}
}
private static void removeMateInfo(SAMRecord rec) {
if (rec.getReadPairedFlag()) {
// Remove all information of its mate
// flag
rec.setProperPairFlag(false); // not paired any more
rec.setMateUnmappedFlag(false);
rec.setMateNegativeStrandFlag(false);
// entries
rec.setMateReferenceIndex(SAMRecord.NO_ALIGNMENT_REFERENCE_INDEX);
rec.setMateAlignmentStart(0);
rec.setInferredInsertSize(0);
// TODO: remove tags and values that are mate pair inclined.
}
}
private void collectReadData(final SAMRecord record, final ReferenceSequence ref) {
metrics.TOTAL_READS++;
readLengthHistogram.increment(record.getReadBases().length);
if (!record.getReadFailsVendorQualityCheckFlag()) {
metrics.PF_READS++;
if (isNoiseRead(record)) metrics.PF_NOISE_READS++;
if (record.getReadUnmappedFlag()) {
// If the read is unmapped see if it's adapter sequence
final byte[] readBases = record.getReadBases();
if (!(record instanceof BAMRecord)) StringUtil.toUpperCase(readBases);
if (isAdapterSequence(readBases)) {
this.adapterReads++;
}
} else if (doRefMetrics) {
metrics.PF_READS_ALIGNED++;
if (!record.getReadNegativeStrandFlag()) numPositiveStrand++;
if (record.getReadPairedFlag() && !record.getMateUnmappedFlag()) {
metrics.READS_ALIGNED_IN_PAIRS++;
// Check that both ends have mapq > minimum
final Integer mateMq = record.getIntegerAttribute("MQ");
if (mateMq == null
|| mateMq >= MAPPING_QUALITY_THRESOLD
&& record.getMappingQuality() >= MAPPING_QUALITY_THRESOLD) {
++this.chimerasDenominator;
// With both reads mapped we can see if this pair is chimeric
if (Math.abs(record.getInferredInsertSize()) > maxInsertSize
|| !record.getReferenceIndex().equals(record.getMateReferenceIndex())) {
++this.chimeras;
}
}
}
}
}
}
