From jisca.huismaned.ac.uk  Sun Feb 23 10:54:15 2014
From: Jisca Huisman <jisca.huismaned.ac.uk>
To: Multiple Recipients of <crimap-usersanimalgenome.org>
Subject: When to stop flipping?
Date: Sun, 23 Feb 2014 10:54:15 -0600

Dear all,

Could anyone please advice me on when to stop doing flips, based on their
experience or guidelines? We have a reasonably good idea of the
order of our SNPs based on a related species, but besides some known  
major rearrangements most likely many minor rearrangements have occurred. 

Within each round of flips, which threshold to use to try and flip? So far
we have rather arbitrarily used a relative log likelihood of -0.5, but I
would like to know where the threshold lies between exploring enough
alternative maps, versus the risk of creating 'false positives'. Sometimes
after making one flip with a low relative logL (e.g. -0.6), several other
flips with much higher relative logL are suggested (e.g. -4 or -5). Using a
fixed threshold would be helpful, both in terms of reproducibility and to
automate the process.

Additionally, is there a way to predict if flips(n+1) is worth doing? For
example, if flips3 gives little improvement in logLikelihood, is it
worthwhile to do the much slower flips4? And how many 'alternative
scenarios' should one consider, where one starts of with a different
initial set of flips?

Thank you,

Jisca Huisman, Postdoctoral researcher Wild Evolutionary Genetics group
Institute of Evolutionary Biology University of Edinburgh

 

--