Do yourself a favour and repeat the alignment to the entire genome. Alignment to subsets leads to false positives because capture-based techniques always have off-targets outside of the capture on-target regions, but as these are not in the reduced genome the aligner will still try to align them somewhere, even if this is not the true origin of the sequence. It is good practice to align to the entire genome, there are multiple threads on BIostars discussing this, plus it saves you from the effort to do any custom extrapolation back to the entire genome. You will see from the alignment stats that a notable portion of reads will align outside of the targets, that is simply how it is, noise/off-targets are inherent in any assay.