Membrane Proteins: Database, Topology, Prediction methods

David Jones jones at
Fri Sep 19 07:31:04 EST 1997

In bionet.molec-model Christian.Lacombe at wrote:
> I am looking for a recent "Membrane Proteins Database" similar to the
> TMbase built by Kay Hofman at Lausanne (Switzerland). As, in this database
> entries are from the SwissProt25 (1994?), I easily imagine that the
> membrane spanning domains (when predicted) are not predicted with
> up-to-date prediction methods. I need also topology information.

The transmembrane records in SWISSPROT are very unreliable. Quite a few
recent papers on topology prediction have made use of the 68 or so proteins
I collected when developing MEMSAT [Biochemistry 33:3038-3049(1994)]. Back
then I had to get most of the information for those proteins from the
literature along with some careful by-eye studies of the sequences and their
hydrophobicity plots.

We are now developing an updated version of MEMSAT, and are in the process
of expanding the original dataset. Unfortunately 4-years on it still looks
like the only reliable option is to plough through the literature - though
at least there are a now couple of new 3-D structures available! If anyone
has a more up-to-date collection then I'd certainly like to hear about it

> It seems that surface amphipatic small helices (the so-called "fusion
> helices"?) are not included in TMbase: how such  membrane segments are
> detected since algorithms predicted only transmembrane segments? And
> finally, how distinguish membrane sheets and membrane helices with
> prediction method?

Best advice at the moment is to model porin-homologue proteins by careful
comparative modelling - using a sequence profile of some description.
until we have more than one example of beta-sheet containing TM proteins
there little else that's sensible to do. Unlike helical bundles, you can't
just predict strands individually and hope to make sense of the results -
unless you know your protein looks like porin, you have no way of knowing
how the strands should be arranged in 3-D space. However, both
transmembrane strands (in porin) and amphipathic helices are amenable to
standard globular protein prediction methods. For helical proteins, the
best strategy would be to carefully predict the TM helices, then apply any
half-decent secondary structure prediction method (even a helical wheel
plot will do) to any long connecting loops you end up with. If the connecting
loops are very long, then you can expect that this is probably an inserted
globular domain (so all bets are off for those regions).

Oh and watch out for the signal peptides...

This message was written, produced and executively directed by Dr David Jones
Address: Dept. of Biological       |   Email: jones at
Sciences, University of Warwick,   |     Tel: +44 1203 523729
Coventry CV4 7AL, U.K.             |     Fax: +44 1203 523568

More information about the Proteins mailing list