[RASMB] Sedimentation Velocity Experiments that are hard to fit

[David Hayes]

Hi to all experienced Ultracentrifuge Scientists,

I have some data (repeated experiments) from sample A buffer 1 (which may 
or may not have been produced on the planet earth -- that is all I can say 
about it)

There are two puzzling aspects to this data.
First, I am not getting good repeatability in the amount of aggregate.
Second, by eye, the raw data looks like a single species (which it should 
be) with a small amount of higher weight aggregate (not surprising), but 
the entire data set never fits well no matter what I try.
Thinking that the repeatability problem might be an artifact of my 
inexperience using Sedfit I tried many different strategies:  including 
floating or not floating just about every parameter and also using the 
nifty "experimental initial distribution" option of Sedfit.
   Fitting to the whole data set, I got a fit with a single large peak and 
two very small aggregate bumps which all together integrate to about 4% of 
the total.  The rmsd was a bit large at 0.01 and the residuals were 
systematic.  Early scans were fit very poorly and later scans fit somewhat 
poorly with scans in the middle fit pretty well.  Later using Sedfit with 
the single species model and only the later half of the scans and floating 
a lot of things and turning off the hat function in Sedfit brought me down 
to a comparable fit with a rmsd of .006, but the residuals were still 
systematic and the molecular weight was off more than what I expected for 
having 4% aggregate.
Then I did something that Peter Shuck thinks is a bit strange, but I think 
it helped me understand the data a bit more.  I fit pairs of scans with 
c(s) taking scans at different times paired with the last 
scan.  Essentially then, I was fitting each scan individually:  using the 
last scan in ever pair was needed to help Sedfit know the baseline.  Peter 
reminded me that by fitting each scan, I lose the ability to distinguish 
heterogeneity from diffusion.  But in this case, all the evidence points to 
a single species, so I did not think heterogeneity (beyond the measured 4% 
aggregation) was significant.  The peak S value stays constant all the way 
down the cell.  Each individual scan fit this way fit very well without 
systematic residuals and a rmsd of about .005.  What changed was that early 
scans fit to an f/f0 of 1.06 while by the later scans f/f0 fit to 
1.5.  This means that the early scans are broader than they should be for a 
reasonable f/f0 and that the later scans seem to be narrower and diffusing 
less.
I repeated this with Sedanal dcdt based curve fitting (here I took scans by 
twos going down the cell and fit to a single species ignoring the 4% 
aggregate) and found a similar trend:  the fitted molecular weight went 
from 95,000 to 225,000 and then settled down to about 160,000.  If there 
were heterogeneity, the scans would fit the opposite way to a one species 
model, the f/f0 would go down, or the weight would go up later in the run 
as the species sedimented apart.

I had some interference data of Tropomyosin that I subjected to a similar 
type of analysis with Sedanal, fitting the whole data set and then fitting 
scans by sets of 10 for the different times in the experiment (because of 
TI, RI noise fitting, it is not possible to fit small sets of scans with 
Sedfit using interference data; however, a Sedfit fit to the whole Tm data 
set showed no systematic variation of the residuals dependent on scan 
number).  There was no scan number dependent trend in molecular weight with 
this molecule, though the variability in fitted molecular weight was 
surprising (weights from different sets of 10 varied from 60,000 to 80,000 
randomly).

Seeing this I tried a Sedfit single species and Sedanal single species fits 
with non-ideality parameters, but floating these parameters made no 
substantial difference.

I don't know if anyone has run into this type of problem, but I am out of 
ideas what to try next.  I am being a bit stubborn wanting a good fit to 
the data:  the S value of the main peak is reproducible and the data 
contains quite a bit of information, but I can't figure out exactly what is 
going on:

Is this just non-ideality that I am not fitting properly or maybe that is 
not modeled well?
Is the culprit convection?  Thoughts on convection in the next message.

Cheers



Dr. David B Hayes
Boston Biomedical Research Institute
64 Grove St.
Watertown, MA  02472
617-658-7738

Boston Biomedical Research Institute... Today's Research for Tomorrow's 
Health.
Please visit us at www.bbri.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.rasmb.org/pipermail/rasmb-rasmb.org/attachments/20060324/4e3e87ca/attachment.htm>