Hi. I have to say I'm embarrassed by this one, but then, that's life.
The error has been in the code since the dawn of time, it seems!

On 03/02/2017 09:28, Geoff Barton wrote:
> BLOSUM62 should be symmetrical, so that is is a typo.
[sic] :)

Mungo - may I suggest that rather than simply fixing the typo, we:
1. Provide a legacy option allow the broken BLOSUM62 matrix to be used
in place of the real matrix. (see, e.g.
https://issues.jalview.org/browse/JAL-728 )

2. Move the score matrices out of static initialisers into flat files
under resources, either as the NCBI matrix format or the aaindex format:
http://www.genome.jp/dbget-bin/www_bget?aaindex:HENS920102

(there might not be an issue about importing AAindex matrices yet..)

> On 03/02/2017 09:17, Mungo Carstairs (Staff) wrote:
>> Is this a typo? I thought pairwise scoring has to be symmetric (PAM250
>> matrix is).
Not sure that holds in general (there's no requirement in biology for
it, and most alignment methods don't depend on it).

Log odds matrices as formulated by Henikoff & Henikoff are symmetric,
however.

>> The help page for PCA says that the Jalview mode calculation (e.g.
>> using BLOSUM62) creates an asymmetric score matrix. Is that correct,
I guess that's just a knock on effect.

However, it's not really relevant - the 'Jalview mode' is the variant of
the seqspace algorithm first provided by Jalview.

I discovered we were doing something different to the original method
back in 2012:
https://issues.jalview.org/browse/JAL-1125

>> or is it actually symmetric (if the score matrix also is)?

perhaps you can verify ?

Thanks again for your eagle-eyed exhaustive testing, Mungo!
Jim.


-- 
-------------------------------------------------------------------
Dr JB Procter, Jalview Coordinator, The Barton Group
Division of Computational Biology, School of Life Sciences
University of Dundee, Dundee DD1 5EH, UK.
+44 1382 388734 | www.jalview.org | www.compbio.dundee.ac.uk

_______________________________________________
Jalview-dev mailing list
[email protected]
http://www.compbio.dundee.ac.uk/mailman/listinfo/jalview-dev

Reply via email to