Modelling protein structures

bionet at cgmvax.cgm.cnrs-gif.fr bionet at cgmvax.cgm.cnrs-gif.fr
Thu Feb 2 04:03:47 EST 1995


Hi,

>: >
>: >This is much too pesimistic. About one third of all currently known
>: >sequences are related to at least one currently known structure.
>: >
>
>: ??? You really mean that 15,000 sequences from Swissprot (for example) are
>: related to at least one entry in the PDB ? I'd be interested in getting a
>: reference on this subject.
>

>
>There are plenty of references on the subject, they basicly stem from the
>fact that the number of different folds seems to be relatively low (estimates
>vary from 100 to 1000).
>

Yes, yes (in fact, rather 1000 than 100). But are you speaking of FOLDS or
SEQUENCES? It is certainly well known that completely different sequences
can produce remarkably similar structures. So, when you say " About one
third of all currently known sequences are related to at least one
currently known structure" maybe you mean that "1/3 of currently known
sequences have a FOLD similar to some one in the PDB". Here I agree. But,
unfortunately, in the absence of clear sequence similarity, this is not of
great help for modelling... 

>
>Of course ther is a problem in the vias of the PDB, but it seems clear now that
>protein modeling by homology has a very strong 'raison d'etre'.
>

Certainly. But you will agree that when the sequence similarities are weak, 
one must be VERY cautious and refrain from overoptimism. It's really too 
easy to produce a "model" when it can't be checked by any experimental 
data. You certainly know that even when you've got the diffraction data for 
"native" crystals -but no heavy atom derivative- things are not that 
simple. Just because THEN you can compare your model with REAL data.......

Salud

Jean-Loup






More information about the Bio-soft mailing list