[Bioperl-l] automatic ortholog detection

Maido Remm mremm@ut.ee
Fri, 28 Sep 2001 13:19:57 +0200


> -----Original Message-----
> From: bioperl-l-admin@bioperl.org [mailto:bioperl-l-admin@bioperl.org]On
> Behalf Of Elia Stupka
> Sent: 28. september 2001. a. 12:56
> To: Ewan Birney
> Cc: ensdev; Bioperl; Fugu Project mailing list
> Subject: [Bioperl-l] Comparative genomics
>
>
> Hello,
>
>    during my week in Singapore we had some interesting discussions over
> genome comparisons. The take home messages seemed to be:
>
> 1-one cannot do anything serious about orthologue/homologue/paralog
> detection with sequence similarity alone
>
> 2-To get good scientific results (read research project) one needs one or
> more researchers actually going down there and analysing families with
> the help of phylogenetic trees, and spending considerable time and brain
> on figuring out what is what.

During my 2-year postdoc in Erik Sonnhammer's group we developed 2
alternative fully automatic methods for ortholog detection.
One is using mutiple alignments and trees, the other pairwise BLAST scores
and heuristic clustering algorithm. I prefer the latter method because it
avoids potential errors related to automatic multiple alignment of
eukaryotic proteins. Also the tree building methods give very variable
results depending on parameters used. The correct way to do it with trees
would be to try different tree building methods with different parameters.
The pairwise BLAST score method, that I was working on, is available from
http://www.cgb.ki.se/inparanoid/ . There is also a database of precalculated
orthologs between human and some of the model organisms. The paper
describing the InParanoid approach should be out soon in JMB. If you are
interested in the other  method exploiting multiple alignments and trees,
please contact Erik (Erik.Sonnhammer@cgb.ki.se).

Maido Remm
Estonian Biocentre,
mremm@ut.ee