Figure 5.

Understanding and detecting ghost introgression. (A) A scenario of ghost introgression from an unsampled outgroup lineage, X, into P1a. (B) When ghost introgression has occurred and a quartet including P1a is sampled, introgression may be erroneously inferred between P2 and P3. This occurs because at some introgressed loci P1a will be more distantly related to both P2 and P3, leading to an excess of discordant trees with P2 and P3 sister to one another (top). If instead a quartet including P1b is sampled, there should no longer be an excess of discordant trees (bottom). (C) Ghost introgression should also be detectable via a change (or a lack of change) in branch lengths. True introgression between P2 and P3 should cause them to be more similar; i.e., shorter branch lengths separating them in discordant trees. In contrast, ghost introgression will not make them more closely related in discordant trees than in concordant trees on average. Similarly, the distance between P1a and all ingroup lineages will be higher when it is the recipient of ghost introgression from an outgroup.