/**
* 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;
}
