I was following a reference chain and came across the PhD thesis of
Jonathan Laserson, who was a Daphne Koller student and now works at Google.
The thesis is interesting. It's quite ambitious: in one chapter he sets up a Bayesian inference algorithm for a tree of cells (cc @mathmomike) that explicitly models sequencing error. On the other hand, I can't see how he got any of this to work efficiently for the sizes of data sets discussed, and his software, called ImmuniTree, is not to be found on the web.
@jeetsukumaran and others may know this fellow from his Genovo de novo assembly software for mixed populations.