[RASMB] problem with fitting

[Peter Schuck]

Hi Marina and Walter ,

> Second SEDFIT does not have a monomer-hexamer model that can be used 
> for direct fitting to a monomer-hexamer model.
>
For years now, the explicit interaction models for self-association, 
binary and ternary associating systems have been moved from SEDFIT to 
SEDPHAT, which can be started seamlessly from within SEDFIT, funneling 
the data along from one into the other.  (Only a few self-association 
models remain in SEDFIT duplicating those in SEDPHAT.)  This is probably 
the easiest for any SEDFIT user, since SEDPHAT has a very similar user 
interface and is based on the same principles.  I think nobody will have 
a problem finding the monomer-hexamer model as a special case of the 
monomer-n-mer model with n = 6.
> If the material is pure and exhibits no concentration dependence you 
> can use SEDFIT c(s) and C(M) to get the molecular weight.
>
> However if it is reversibly associating/dissociating, you will need to 
> use other software, like SEDANAL, in which you can specify a model of 
> your choice.
> Of course, you can use SEDANAL also to get the molar mass of 
> non-interacting species as well without making any assumptions about 
> frictional ratios.
>
Sorry for everybody remembering this discussion, I know this is tedious 
because it has been said many times before:  What one needs for c(M) to 
get good molecular weight estimates (usually within 5-10% of the true 
molar mass or better, assuming that vbar and buffer density are known) 
is not that the sample is pure - but only that the species of interest 
is the main peak.  If that's not the case, one may use, for example, the 
bimodal frictional ratio model.  Second, it does still work well 
(certainly sufficient to distinguish trimer from hexamer) for reacting 
systems if the species of interest is kinetically stable because of 
either a slow off-rate constant (koff < 1e-4) or because it is kept 
populated via mass action law by having a concentration >> Kd or << Kd.  
Both conditions can be checked by confirming that the c(s) peak of 
interest does not show a concentration dependence (when trying, for 
example, 3fold higher and 3fold lower concentrations). 

An essential criterion for any c(s) or c(M) analysis, or any other 
analysis for that matter, is that the model actually fits the raw data.

What Walter proposes to do in his last sentence is modeling with a 
discrete Lamm equation solution or variation of that.  Note that 
'without making any assumptions about frictional ratios' in SEDANAL 
means without diffusional deconvolution.  As is well-published, 
attempting to determine molar masses without resolving possible 
heterogeneity contributions to the boundary spread can give very poor 
results, frequently vastly underestimating the molar mass, even in the 
absence of chemical reactions.  The discrete approach actually requires 
much higher purity than the c(M) approach.  This can be easily tested 
within SEDFIT or SEDPHAT, which incorporates both approaches and hybrids 
thereof. Usually the discrete Lamm equation solutions give too low Mw's, 
much higher rmsd and clearly systematic residual bitmaps.

To come back to Marina's data - I don't know where the discrepancy in 
the obtained molar mass from the expected values comes from, but I don't 
think it can be fixed by using a discrete Lamm equation solution.  I 
would suggest to double-check in a concentration series that the c(s) 
peak does not change position with concentration.  Studying this problem 
with sedimentation equilibrium will have the same requirement of using a 
concentration series making sure that the highest species remains 
sufficiently populated (via mass action law), otherwise the result will 
also be an average over the populations seen in the radial range of data 
analysis.

Also, more basic, I think it is important to look carefully at the 
temperature equilibration (in particular with the run being done at 8 
C), and the effect of temperature and buffer composition on density and 
viscosity.  And as I said above, it is essential that the fit is good, 
otherwise this indicates that something is wrong and one can't trust 
very much the numbers coming from the fit.

Best regards,
Peter Schuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.rasmb.org/pipermail/rasmb-rasmb.org/attachments/20080311/9bc21256/attachment.htm>