[RASMB] Low s value peaks

[Peter Schuck]

Samantha,

While writing the following response, addressing Jack Correia's skepticism, 
John Philo had already stated what I wanted to say about your problem.

Jack, you won't be surprised if I don't  agree with your view.  In your 
truncated description of c(s), you are ignoring several things, like the 
regularization, non-negativity constraints, and most of all the fact that 
Lamm equation solutions with different s-values are simply not much 
correlated.  To understand the method, these considerations are actually 
quite important, because they make it in general a well-conditioned 
problem.   But of course every method has its limits, which are important 
to consider (and which have been explored and described in detail).

Also, if one has more information, then of course it's always good to put 
it in.  This includes the prior assumption that there are actually a few 
discrete species present.  There's also an intermediate approach possible, 
for example, in SEDPHAT to use a hybrid c(s)/discrete species model (see
http://www.analyticalultracentrifugation.com/sedphat/hybrid_local_continuos_global_discrete.htm
) to replace peak regions of c(s) with distinct species that have molar 
mass values that can be either used from prior knowledge, or fitted 
(including constraining them to be multiples of monomers for discrete 
oligomers), in combination with piecewise continuous c(s) sections for the 
unknown species.

But I believe these issues have actually not much to do with Samantha's 
problem.  I completely agree with John Philo here.  In my experience, peaks 
at small s-values can be caused by two things:

1) an imperfectly matched buffer reference, such that there's a small 
sedimenting buffer component in excess in the sample versus reference 
compartment.  If that's the case, e.g., in your case if one would leave out 
the azide or DTT from the reference buffer, then this can sometimes be very 
clearly and reproducibly be picked up as a peak in c(s).

[In this case, I would actually suggest to proceed with a slighly different 
model.  c(s) with prior knowledge, either as a bimodal c(s) or as c(s) with 
one discrete species can be used to describe the sedimentation of the 
buffer components separately, uncoupled from the macromolecules you're 
interested in.  These models are described in
http://www.analyticalultracentrifugation.com/sedfit_help_cs_with_other_prior.htm
For example the model with the extra small component has worked well for us.]

2) more frequently encountered is the situation when the smallest s-value 
chosen is so small that under the conditions of the experiment it simply 
cannot be distinguished from a constant baseline.  This depends on the 
actual rotor speed, sets of scans considered, fitting limits, and the 
smallest s-value you considered as s-min in c(s).  If this is the problem, 
there's nothing to worry about, as the similarity of a very tiny 
sedimentation and no sedimentation (i.e. baseline) just will make the 
amplitude of the smallest s-values correlated with the baseline 
parameter.   What Jack mentioned, correlation between parameters does 
happen here, but it is limited to the baseline and s-min parameters, and 
there will be no correlation with higher s-values.  Your c(s) curves at 
values > smin should look exactly the same.  Therefore this is not a 
problem (unless one would be primarily interested in the baseline or 0.1 S 
species).

There is a simple way to distinguish the two cases.  If (2) is true, then 
setting the s-min value to a slightly higher value will decrease the height 
of the c(s) at smin, and eventually makes the apparent peak disappear 
without changing the rmsd of the fit.  If (1) were true, then moving s-min 
to a higher value would decrease the quality of the fit, because an 
observable sedimenting component would be neglected.


Peter Schuck



At 11:49 AM 1/23/2004 -0600, you wrote:
>----------------------------------------------------------------------------------
>The older archived RASMB emails can be found at:
>http://rasmb-email.bbri.org/rasmb_archives
>and current archives at
>http://rasmb-email.bbri.org/pipermail/rasmb/
>Search All the Archives at:
>http://rasmb-email.bbri.org/rasmb_search.html
>----------------------------------------------------------------------------------
>
>Peaks in C(s) do not necessarily mean that it is an actual peak in the
>sample - spurious peaks or shifting peak positions in sedfit is caused
>by the fact you are fitting the data to linear combinations of 100 s,
>common f/fo values. This is remarkable overdetermined and the actual
>amplitudes you get are not necessarilly reliable. If they are not also
>supported by a noninteracting species fit then they very well may be due
>to the nature of a c(s) fit.
>
>Yes, sedfit can sensitively pick up small zones, but I never trust them
>unless I repeatedly see it in multiple samples and if possible can
>verify their presence by direct boundary fitting to a more realistic
>model, ie fewer parameters.
>
>
>-------------------------------------------------------------------
>  Dr. John J. "Jack" Correia
>  Department of Biochemistry
>  University of Mississippi Medical Center
>  2500 North State Street
>  Jackson, MS 39216
>  (601) 984-1522
>  fax (601) 984-1501
>  email address: jcorreia at biochem.umsmed.edu
>  homepage location: http://biochemistry.umc.edu/correia.html
>  dept homepage location:   http://biochemistry.umc.edu/
>-------------------------------------------------------------------
>
>
>
>
> >>> "Jones, Samantha" <samantha.jones at imperial.ac.uk> 01/23/04 11:20AM
> >>>
>Hello all
>I have been trying to analyse some interference data on Sedfit and
>have
>been repeatedly seeing a large 'half' peak at very low s values on my
>s
>value distributions.
>On the continuous c(s) distributions after fitting to range of 0.1s to
>10s the distribution line starts at c(s) values of 1.5 to 0.5 on 0 on
>the x axis, dropping to 0 on the y axis by 0.1s value. This is still
>the
>case after changing the parameters to 0.01s to 10s.
>The buffer used at first was 10mM NaAcetate, 0.2% Azide and 2mM DTT pH
>4. At first I thought it could be the DTT but I saw the same thing
>using
>a sample in a buffer 10mM Tris 10mM Na Acetate, 0.2% Azide no DTT pH 8.
>
>I have seen this spinning at 30K and 50K for about 16hours.
>
>Could Azide be the problem? I don't expect either sample to be
>contaminated, however anything of this size would have been missed by
>the SDS-PAGE gel run on both samples.
>
>Thank you
>
>Samantha Jones
>Samantha Jones
>MRC Prion Unit
>St Marys Hospital
>Norfolk Place
>London
>W2 1PG
>
>_______________________________________________
>RASMB mailing list
>RASMB at rasmb-email.bbri.org
>http://rasmb-email.bbri.org/mailman/listinfo/rasmb