William R. Pearson <wrp at alpha0.bioch.virginia.edu> wrote:
> I strongly disagree with this statement. With the exceptions of
> extremely biased sequences, like histones and low complexity regions,
> I know of no examples where "high sequence similarity" is misleading.
> In particular, all transmembrane helices do not look remarkably alike
> from the Blast/SW/Fasta perspective. We routinely use various
> receptors with multiple transmembrane domains to test the FASTA
> package, and we never see statistically significant similarity to
> unrelated membrane proteins. Dealing with highly biased sequence
> composition is an area of ongoing research, the latest version of the
> FASTA package (33t06) offers a statistical estimation option (-z 6)
> that includes a composition bias correction.
>
Although I'm not going to argue with what Bill says here - it might be
worth noting that the above comment applies to "normal" Blast only.
Using iterated Blast (PSI-BLAST) with transmembrane proteins can produce
_apparently_ very significant matches to completely unrelated transmembrane
proteins after only two iterations. Given that PSI-BLAST is often used as
a black box for genome annotation, this is a serious problem. The underlying
problem is of course that in BLAST the E-value estimates are based on a
certain set of assumptions - even more assumptions in PSI-BLAST - and these
are not applicable in certain common cases.
Although in my lab we try to avoid using PSI-BLAST with transmembrane
proteins as far as possible - when we do use it for these cases we use a
much lower iteration E-value cutoff (1e-16 as opposed to the default 1e-3).
But, even this is not guaranteed foolproof.
- David -
--
>-----------------------------------------------------------------------------<
This text was written, produced and executively directed by Prof. David Jones
Address: Dept. of Biological | Email: David.Jones at brunel.ac.uk
Sciences, Brunel University, | Tel: +44 1895 274000 x3942 /x3984 (Sec.)
Uxbridge, Middlesex UB8 3PH, U.K. | Fax: +44 1895 274348
