I've seen two uses of this term that seem to refer to completely different phenomena. Is my understanding correct?
- In Jain et al. "Active promoters give rise to false positive ‘Phantom Peaks’ in ChIP-seq experiments" (PMC4538825), a phantom peak is a region on the genome where lots of reads align in a control experiment where there is none of the protein being IP'd, but not in a control experiment without any antibody. The authors (speculatively) attribute this to general stickiness of transcription factors and the transcriptional machinery, which can bind with potentially any antibody and get IP'd.
- The ENCODE quality metrics (link) say "a cross-correlation phantom-peak is ... observed at a strand-shift equal to the read length." They are referring to a high correlation of x[i] and y[i + r] where r is the read length, x[i] contains counts of tags aligning to the Watson strand at coordinate i, and y[i+r] contains counts of tags aligning to the Crick strand at coordinate i+r. They attribute this phenomenon to variations in mappability. (I don't understand their explanation at all, but that's a separate question.)