private static String createVerboseOutput(final ReadBackedPileup pileup) {
final StringBuilder sb = new StringBuilder();
boolean isFirst = true;
sb.append(pileup.getNumberOfDeletions());
sb.append(" ");
for (PileupElement p : pileup) {
if (isFirst) isFirst = false;
else sb.append(",");
sb.append(p.getRead().getReadName());
sb.append(verboseDelimiter);
sb.append(p.getOffset());
sb.append(verboseDelimiter);
sb.append(p.getRead().getReadLength());
sb.append(verboseDelimiter);
sb.append(p.getRead().getMappingQuality());
}
return sb.toString();
}
private Haplotype getHaplotypeFromRead(
final PileupElement p, final int contextSize, final int locus) {
final GATKSAMRecord read = p.getRead();
int readOffsetFromPileup = p.getOffset();
final byte[] haplotypeBases = new byte[contextSize];
Arrays.fill(haplotypeBases, (byte) REGEXP_WILDCARD);
final double[] baseQualities = new double[contextSize];
Arrays.fill(baseQualities, 0.0);
byte[] readBases = read.getReadBases();
readBases =
AlignmentUtils.readToAlignmentByteArray(
read.getCigar(), readBases); // Adjust the read bases based on the Cigar string
byte[] readQuals = read.getBaseQualities();
readQuals =
AlignmentUtils.readToAlignmentByteArray(
read.getCigar(),
readQuals); // Shift the location of the qual scores based on the Cigar string
readOffsetFromPileup =
AlignmentUtils.calcAlignmentByteArrayOffset(
read.getCigar(), p, read.getAlignmentStart(), locus);
final int baseOffsetStart = readOffsetFromPileup - (contextSize - 1) / 2;
for (int i = 0; i < contextSize; i++) {
final int baseOffset = i + baseOffsetStart;
if (baseOffset < 0) {
continue;
}
if (baseOffset >= readBases.length) {
break;
}
if (readQuals[baseOffset] == PileupElement.DELETION_BASE) {
readQuals[baseOffset] = PileupElement.DELETION_QUAL;
}
if (!BaseUtils.isRegularBase(readBases[baseOffset])) {
readBases[baseOffset] = (byte) REGEXP_WILDCARD;
readQuals[baseOffset] = (byte) 0;
} // N's shouldn't be treated as distinct bases
readQuals[baseOffset] = (byte) Math.min((int) readQuals[baseOffset], p.getMappingQual());
if (((int) readQuals[baseOffset]) < 5) {
readQuals[baseOffset] = (byte) 0;
} // quals less than 5 are used as codes and don't have actual probabilistic meaning behind
// them
haplotypeBases[i] = readBases[baseOffset];
baseQualities[i] = (double) readQuals[baseOffset];
}
return new Haplotype(haplotypeBases, baseQualities);
}
@Override
protected Double getElementForPileupElement(final PileupElement p) {
return (double) p.getRead().getMappingQuality();
}
@Override
public T traverse(
final ActiveRegionWalker<M, T> walker, final LocusShardDataProvider dataProvider, T sum) {
logger.debug(String.format("TraverseActiveRegion.traverse: Shard is %s", dataProvider));
final LocusView locusView = getLocusView(walker, dataProvider);
final GenomeLocSortedSet initialIntervals = engine.getIntervals();
final LocusReferenceView referenceView = new LocusReferenceView(walker, dataProvider);
final int activeRegionExtension =
walker.getClass().getAnnotation(ActiveRegionExtension.class).extension();
final int maxRegionSize =
walker.getClass().getAnnotation(ActiveRegionExtension.class).maxRegion();
if (locusView
.hasNext()) { // trivial optimization to avoid unnecessary processing when there's nothing
// here at all
int minStart = Integer.MAX_VALUE;
ActivityProfile profile =
new ActivityProfile(engine.getGenomeLocParser(), walker.hasPresetActiveRegions());
ReferenceOrderedView referenceOrderedDataView =
getReferenceOrderedView(walker, dataProvider, locusView);
// We keep processing while the next reference location is within the interval
GenomeLoc prevLoc = null;
while (locusView.hasNext()) {
final AlignmentContext locus = locusView.next();
GenomeLoc location = locus.getLocation();
if (prevLoc != null) {
// fill in the active / inactive labels from the stop of the previous location to the
// start of this location
// TODO refactor to separate function
for (int iii = prevLoc.getStop() + 1; iii < location.getStart(); iii++) {
final GenomeLoc fakeLoc =
engine.getGenomeLocParser().createGenomeLoc(prevLoc.getContig(), iii, iii);
if (initialIntervals == null || initialIntervals.overlaps(fakeLoc)) {
profile.add(
fakeLoc,
new ActivityProfileResult(
walker.hasPresetActiveRegions()
&& walker.presetActiveRegions.overlaps(fakeLoc)
? 1.0
: 0.0));
}
}
}
dataProvider.getShard().getReadMetrics().incrementNumIterations();
// create reference context. Note that if we have a pileup of "extended events", the context
// will
// hold the (longest) stretch of deleted reference bases (if deletions are present in the
// pileup).
final ReferenceContext refContext = referenceView.getReferenceContext(location);
// Iterate forward to get all reference ordered data covering this location
final RefMetaDataTracker tracker =
referenceOrderedDataView.getReferenceOrderedDataAtLocus(
locus.getLocation(), refContext);
// Call the walkers isActive function for this locus and add them to the list to be
// integrated later
if (initialIntervals == null || initialIntervals.overlaps(location)) {
profile.add(location, walkerActiveProb(walker, tracker, refContext, locus, location));
}
// Grab all the previously unseen reads from this pileup and add them to the massive read
// list
for (final PileupElement p : locus.getBasePileup()) {
final GATKSAMRecord read = p.getRead();
if (!myReads.contains(read)) {
myReads.add(read);
}
// If this is the last pileup for this shard calculate the minimum alignment start so that
// we know
// which active regions in the work queue are now safe to process
minStart = Math.min(minStart, read.getAlignmentStart());
}
prevLoc = location;
printProgress(locus.getLocation());
}
updateCumulativeMetrics(dataProvider.getShard());
// Take the individual isActive calls and integrate them into contiguous active regions and
// add these blocks of work to the work queue
// band-pass filter the list of isActive probabilities and turn into active regions
final ActivityProfile bandPassFiltered = profile.bandPassFilter();
final List<ActiveRegion> activeRegions =
bandPassFiltered.createActiveRegions(activeRegionExtension, maxRegionSize);
// add active regions to queue of regions to process
// first check if can merge active regions over shard boundaries
if (!activeRegions.isEmpty()) {
if (!workQueue.isEmpty()) {
final ActiveRegion last = workQueue.getLast();
final ActiveRegion first = activeRegions.get(0);
if (last.isActive == first.isActive
&& last.getLocation().contiguousP(first.getLocation())
&& last.getLocation().size() + first.getLocation().size() <= maxRegionSize) {
workQueue.removeLast();
activeRegions.remove(first);
workQueue.add(
new ActiveRegion(
last.getLocation().union(first.getLocation()),
first.isActive,
this.engine.getGenomeLocParser(),
activeRegionExtension));
}
}
workQueue.addAll(activeRegions);
}
logger.debug(
"Integrated "
+ profile.size()
+ " isActive calls into "
+ activeRegions.size()
+ " regions.");
// now go and process all of the active regions
sum = processActiveRegions(walker, sum, minStart, dataProvider.getLocus().getContig());
}
return sum;
}
private double scoreReadAgainstHaplotype(
final PileupElement p, final int contextSize, final Haplotype haplotype, final int locus) {
double expected = 0.0;
double mismatches = 0.0;
// What's the expected mismatch rate under the model that this read is actually sampled from
// this haplotype? Let's assume the consensus base c is a random choice one of A, C, G, or T,
// and that
// the observed base is actually from a c with an error rate e. Since e is the rate at which
// we'd
// see a miscalled c, the expected mismatch rate is really e. So the expected number of
// mismatches
// is just sum_i e_i for i from 1..n for n sites
//
// Now, what's the probabilistic sum of mismatches? Suppose that the base b is equal to c.
// Well, it could
// actually be a miscall in a matching direction, which would happen at a e / 3 rate. If b !=
// c, then
// the chance that it is actually a mismatch is 1 - e, since any of the other 3 options would be
// a mismatch.
// so the probability-weighted mismatch rate is sum_i ( matched ? e_i / 3 : 1 - e_i ) for i = 1
// ... n
final byte[] haplotypeBases = haplotype.getBases();
final GATKSAMRecord read = p.getRead();
byte[] readBases = read.getReadBases();
readBases =
AlignmentUtils.readToAlignmentByteArray(
p.getRead().getCigar(), readBases); // Adjust the read bases based on the Cigar string
byte[] readQuals = read.getBaseQualities();
readQuals =
AlignmentUtils.readToAlignmentByteArray(
p.getRead().getCigar(),
readQuals); // Shift the location of the qual scores based on the Cigar string
int readOffsetFromPileup = p.getOffset();
readOffsetFromPileup =
AlignmentUtils.calcAlignmentByteArrayOffset(
p.getRead().getCigar(), p, read.getAlignmentStart(), locus);
final int baseOffsetStart = readOffsetFromPileup - (contextSize - 1) / 2;
for (int i = 0; i < contextSize; i++) {
final int baseOffset = i + baseOffsetStart;
if (baseOffset < 0) {
continue;
}
if (baseOffset >= readBases.length) {
break;
}
final byte haplotypeBase = haplotypeBases[i];
final byte readBase = readBases[baseOffset];
final boolean matched =
(readBase == haplotypeBase || haplotypeBase == (byte) REGEXP_WILDCARD);
byte qual = readQuals[baseOffset];
if (qual == PileupElement.DELETION_BASE) {
qual = PileupElement.DELETION_QUAL;
} // calcAlignmentByteArrayOffset fills the readQuals array with DELETION_BASE at deletions
qual = (byte) Math.min((int) qual, p.getMappingQual());
if (((int) qual)
>= 5) { // quals less than 5 are used as codes and don't have actual probabilistic meaning
// behind them
final double e = QualityUtils.qualToErrorProb(qual);
expected += e;
mismatches += matched ? e : 1.0 - e / 3.0;
}
// a more sophisticated calculation would include the reference quality, but it's nice to
// actually penalize
// the mismatching of poorly determined regions of the consensus
}
return mismatches - expected;
}