Strategy

For large datasets, doing a thorough analysis of the dataset with TreeLD can be very computationally intensive. Therefore, it is advisable to perform an approximate run as a first pass analysis with a low number of focal points (maybe 1 focal point per 0.1cM), a small number of sampled trees and a low burn-in. This analysis may be sufficient to identify interesting areas in the sequence of interest, which then can be analyzed by a tighter grid of focal points and a more extensive MCMC. The generated trees in this first pass will only be rough approximations of the true ancestry of the locus, but they may already contain some signal about the presence of a disease mutation. If those approximate trees are insufficient and only little signal is generated, treebuilding can be restarted from the last sampled tree and the original trees can be discarded (see 6.6). It should be pointed out that running the program this way should only be done for exploratory analysis of the data as repeated approximate runs generate a multiple-testing problem that can create false positives.

As it is much easier in this version of the program to extend the burn-in or to sample additional trees than it is to add focal points, it may be a sensible approach to plan the analysis accordingly.