Le 17 nov. 09 à 12:40, Morten Kjeldgaard a écrit :
> Tim Gruene wrote:
>
>>>>> Yes, but models that can be validated against experimental data.
>>>>> The
>>>>> defining characteristics of computational models is that they (A)
>>>>> are 100% dependent on the algortihm, (B) can't be validated at
>>>>> all.
>>>>>
>>>>> Cheers,
>>>>> Morten
>>>> Sorry, they can be validated to some extend using biochemical data!
>>>
>>> You are joking, right?
>
>> I would say that any prediction that can be derived from a model and
>> confirmed is a validation of the model and the model remains valid
>> until
>> replaced by a better one. The sun was orbiting the earth until
>> evidence
>> became too contradictorily for this model. Until then it was a good
>> model - better than no model at all, be it wrong or not.
>
> Whoa there. Let's move back a few steps. This discussion started
> because
> someone said that there are "rumblings" that modelbuilding would
> soon be a
> competitive technique to xray-crystallography.
>
That was perhaps a joke. In any case, it can't be seriously
considered. I think the potentially useful discussion is about what
information can we gain from each other.
> I objected with the fact that computational models cannot be
> validated, a
> claim which was countered with "they can be validated using
> biochemical data".
>
> I think that is really funny. So, you want to compute a model of a
> macromolecule from first principles, and then spend the next 10
> years in the
> biochemistry lab validating it? Because that is what it will take
> until you
> can convince anyone that the positions of your loops, your rotamers,
> your
> co-factors and your metal-binding sites are correct.
>
You don't need 10 years to test a clear prediction. From whatever kind
of model. If you need 10 years the prediction is probably useless in
its present form. I presume that is precisely your point. My point is
that such models may produce clear, testable predictions.
From the way we are discussing it would seem that this is a matter of
opinion. The fact is that some models are validated, even
structurally, see:
Qian, B., Raman, S., Das, R., Bradley, P., McCoy, A. J., Read, R. J.,
and Baker, D. (2007) High-resolution structure prediction and the
crystallographic phase problem. Nature, 450: 259–264.
> I thought this list was for crystallographers, but apparently people
> no
> longer understand what "validation" means in structural science.
>
I agree with this, but not in the way you think. Crystallographers may
be/need to be inerested by other fields related to structural biology.
Even if they don't agree with the way other fields research is carried
out.
> (...) OTOH, a poor molecular model may cause
> unlimited waste of time and money by other scientists.
That's what happened for example with the crystal structures of the
Emr multidrug transporters. Biochemists found a hard time to get
funding for research that was in contradiction with those, later
retracted crystallographic models. I hope that they don't conclude
from that episode that trusting crystallographic models is useless or
even dangerous to them.
Best,
-- Miguel
Architecture et Fonction des Macromolécules Biologiques (UMR6098)
CNRS, Universités d'Aix-Marseille I & II
Case 932, 163 Avenue de Luminy, 13288 Marseille cedex 9, France
Tel: +33(0) 491 82 55 93
Fax: +33(0) 491 26 67 20
e-mail: [log in to unmask]
Web: http://www.pangea.org/mol/spip.php?rubrique2
|