Usually, read duplicates, i.e. reads that have exact same start and end positions, are removed before follow-up analysis of sequencing data, since they are supposed to be PCR artifacts. However, in target capture experiments we expect a probably substantial amount of "real" read duplication due to the enrichment process. "Real" read duplicates would be derived from actually separate input DNA molecules that by chance were fragmented at the same position. In the case of paired-end data, the position information of both reads of a pair can be used. Reads only have to be considered duplicated if the positions of both reads are found again in another pair. For read pairs, usually the extent of duplication is by far not as high as for single reads.
The duplicates barplot shows which fraction of read pairs is present in the data in what number of copies. Read multiplicity proportions are calculated and shown separately for on- and off-target read pairs. Therefore, the plot gives an impression about the amount of "real" duplication (expected mostly in the target regions) versus artifactual duplication (expected both on- and off-target). The bar heights represent percentages, absolute numbers are given, in millions, on top of the bars.
> duplicates.barplot(readpairs, targets, ylab="Fraction of read pairs")@DUPLICATES@