On Sun, Mar 25, 2012 at 11:01 PM, Kenneth Nordtvedt <knordtvedt@bresnan.net>wrote: > I'm not sure you want to try > normal MonteCarlo; the tree has to end with the N known present day > haplotypes. Maybe even the world's most super computers could not run > simulations through enough trees from a founder haplotype, so that after > taking the infinitesimal fraction of the simulated trees which ended with > the given final haplotypes could then be examined to see which varieties of > this subset happened most? Although maybe regressing by simulation > backward > in time until tree collapsed to single founder could be worked out? One of us (or both) may have misunderstood the other; I was referring not to a Monte Carlo simulation of the process of going from a founder haplotype to N present day haplotypes, nor to a backward in time convergence from N to 1, but to something like the "simulated annealing" one uses to find the most likely state of a crystal, or a quantum field... or, in this case, of a haplotype tree. But maybe that analogy falls apart on closer examination. I left the connectivity of the tree out to help audience > digestibility before we faced that. What was described in previous message > would have to be done for all alternative tree connectivities in principle, > and the maximum liklihoods for each then compared. N final haplotypes > certainly don't specify the tree connectivity. That would, I think, greatly complicate any attempt at doing this by "annealing". But again, I'm less than expert on both the genetic and the computational technicalities. - Rich Holmes