Hi Elaine.
I was wondering, is there a particular reason Shenkin + normalization is the default AACon setting? Is it "better" or mainly chosen for computational expediency?
Mainly computational expediency - I seem to remember that of the
computationally cheap methods, Shenkin didn't perform very badly, but I
suspect the choice may have been arbitrary. However, read on..
Also, this may be a bugette: when I turned on the AACons annotation with these default parameters set, I found that the result was not normalized. I had to click each option off and then on again, then re-run in order to get the normalized result. (Well, probably I only had to click normalization off and on, but I didn't try that.) See before-and-after screen shots, and I also attach the small alignment.
Again - well spotted ! It's nice having someone to beta test 
There
is indeed a bug at play here - and is now duly lodged:
http://issues.jalview.org/browse/JAL-1532
I'm trying to understand these AACon scores better, but it is somewhat confusing because their ranges (with normalization turned off) are different in some cases than described in the Valdar paper, and I'm not that great at reading equations in the first place. For example, the Valdar score is described as falling in the range 0-1, but I'm guessing the normalization part of the equation is left off in the AACon calculation when the normalization is turned off in the Jalview interface.
Yes. The way AACon works is to perform all the conservation analysis
calculations as efficiently as possible, so the equations were
refactored so that a constant number of summation loops are required to
calculate any combination of scores. Since normalisation requires a
separate loop, it was separated out. Ideally, the documentation would
explain that, of course
Please share if you have any thoughts or recommendations on these scores.
I'm afraid I do not. The first place to look is the paper describing the
SMERF method (The contrasting properties of conservation and correlated phylogeny in protein functional residue prediction - PubMed). Jon Manning
(who did the basic research prior to AACon's implementation) concluded
that Williamson's score was one of the most effective but I'm sure there
are special cases for every one of the methods.
hope that helps - and thanks for spotting the bug!
Jim.
The University of Dundee is a registered Scottish Charity, No: SC015096
ยทยทยท
On 18/07/2014 00:32, Elaine Meng wrote: