Picard AlignmentSummaryMetrics - PCT_CHIMERA definition

[pontushojer]

I am a bit confused as to how chimeras are defined in AlignmentSummaryMetrics.

In the metric definitions (https://broadinstitute.github.io/picard/picard-metric-definitions.html#AlignmentSummaryMetrics) PCT_CHIMERA chimera is defined as:

The fraction of reads that map outside of a maximum insert size (usually 100kb) or that have the two ends mapping to different chromosomes.

In the tool docs (https://gatk.broadinstitute.org/hc/en-us/articles/360040507751-CollectAlignmentSummaryMetrics-Picard) however, it is stated that:

Chimeras are identified if any of the of following criteria are met:

• the insert size is larger than MAX_INSERT_SIZE
• the ends of a pair map to different contigs
• the paired end orientation is different that the expected orientation
• the read contains an SA tag (chimeric alignment)

The tool docs include two additional read types as chimeric, based on pair orientation and those with supplementary alignments. Which one is correct?

[pontushojer]

To add to this, it seems from the code that only reads with a MAPQ >= 20 are counted for this metric, which is not stated in the docs.

picard/src/main/java/picard/analysis/AlignmentSummaryMetricsCollector.java

Line 66 in 6c3f23b

private final int MAPPING_QUALITY_THRESHOLD = 20;

picard/src/main/java/picard/analysis/AlignmentSummaryMetricsCollector.java

Lines 375 to 388 in 6c3f23b

	if (mateMq == null || mateMq >= MAPPING_QUALITY_THRESHOLD && record.getMappingQuality() >= MAPPING_QUALITY_THRESHOLD) {
	++chimerasDenominator;
	// With both reads mapped we can see if this pair is chimeric
	if (ChimeraUtil.isChimeric(record, maxInsertSize, expectedOrientations)) {
	++chimeras;
	}
	}
	} else { // fragment reads or read pairs with one end that maps
	// Consider chimeras that occur *within* the read using the SA tag
	if (record.getMappingQuality() >= MAPPING_QUALITY_THRESHOLD) {
	++chimerasDenominator;
	if (record.getAttribute(SAMTag.SA.toString()) != null) {
	++chimeras;

[kachulis]

The tool docs are (more) correct.
For read-pairs where both reads are mapped, the two reads are counted as chimeric if any of these criteria are met (see :

• the insert size is larger than MAX_INSERT_SIZE
• the ends of a pair map to different contigs
• the paired end orientation is different that the expected orientation

If only one read is mapped (or the reads are single ended), the read is counted as chimeric if:

• the read contains an SA tag (chimeric alignment)

As you also noted, only reads or fragments with mapping qualities above 20 are counted toward either the numerator or denominator when calculating PCT_CHIMERA

[pontushojer]

Thank you @kachulis for the nice breakdown! I guess then the docs should be updated to reflect this definition.