[RASMB] SAUCE - x-ray optics for the AUC

Tue Mar 17 08:55:24 PDT 2009

Borries,

In your last email to RASMB you pointed to your recent paper in Eur. Biophys. J.  on size-and-shape distributions.  I just read it, and since you cite our prior work (Brown & Schuck, Biophys. J. 90 (2006) 4651-4661, http://www.pubmedcentral.nih.gov/articlerender.fcgi?tool=pubmed&pubmedid=16565040) I want to use this opportunity to clarify the points in our work that you BLATANTLY AND FANTASICALLY MISREPRESENTED.  

Our size-and-shape distributions are not dealing with 10x10 grids (number of ff0-values by s-values) as you claimed, but in reality it is 10 x 100 grids or higher.  Everybody can see readily in our paper that we are not using a 10x10 grid - that would not make any sense.  I have really no idea how it could be possible that you thought our work was based on 10x10 grids, especially as you recognize this is a crucial point.  Indeed, all the figures in our work are showing high resolution 2d grids, and we describe them in detail both in the figure legend and repeatedly in the text.  (Strangely, though, reading your paper your own 2DSA method is based on 10 x 10 grids.)

As I am writing this I can't help also commenting on the science.  

As it turns out, your assertion that a 10x10 grid would already require 0.5 GB of RAM correct only for a naïve computational implementation.  Everybody can use SEDFIT and watch the Windows task manager reporting the memory being depleted while doing a c(s,ff0) analysis even with 10x100 is fairly moderate and can be accomplished with standard RAM sizes.  How to do that is published in our papers.

About the computers necessary: The SEDFIT analysis done in the paper was accomplished on a laptop at home with single processor, single thread, taking only a few minutes for each.  Since then, we have implemented some parallel routines in SEDFIT that allow to take advantage of multi-core processors to speed things up quite a bit more.   

Finally, since you describe your 2DSA method as being more 'rigorous', I want to point out that a proper minimization cannot be achieved with what you call a 'divide-and-conquer' approach of fitting a series of (coarse) partial models and then merging the results.  Rigorously, it is a quite simple and a most basic fact of fitting any data:  One needs to have a sufficiently fine model and fit the data properly as a combination of all components simultaneously.  Anything else cannot properly account for all cross-correlations between all species.  This requires a large grid, such as 10x100, and that is what c(s,ff0) method in SEDFIT does.  Another component that is essential to prevent the result from being nonsense is regularization, as is well-known in many scientific disciplines.

Peter

P.S. A point regarding the standard c(s) analysis that is overlooked in your paper is that for the standard c(s) method one can use many different scaling laws and segmented ff0 values as appropriate for the given system.  As it happens the traditional s~M^(2/3) is quite sufficient for working with folded proteins, and therefore the most commonly used approach.  Of course, this would not apply to your model system of a mixture of lysozyme and a nucleic acid, but for such data we would see two boundaries and variation of c(s) with a bimodal frictional ratio would work just fine.

Peter Schuck, PhD
Dynamics of Macromolecular Assembly
Laboratory of Bioengineering and Physical Science, NIBIB, NIH
13 South Drive
Bldg 13, Rm 3N17
Bethesda, MD 20892

phone: (301) 435-1950
fax: (301) 480-1242
email: schuckp at mail.nih.gov

-----Original Message-----
From: rasmb-bounces at rasmb.bbri.org [mailto:rasmb-bounces at rasmb.bbri.org] On Behalf Of Borries Demeler
Sent: Tuesday, March 17, 2009 8:48 AM
To: Arthur Rowe
Cc: rasmb at server1.bbri.org
Subject: Re: [RASMB] SAUCE - x-ray optics for the AUC

> 
> Just a general thought, colleagues - which maybe has been considered
> already. But am I correct in thinking that to get multiple species resolved
> into a form sufficiently 'pure' for SAXS analysis you would be using 'zonal'
> separation i.e. band-forming cells?
> 
> Although few folks seem to actually use this approach in a conventional AUC,
> tit does in fact work well, as does the SEDFIT software for analysis of
> same. You need to avoid working with low MW solutes, of course. Do I guess
> that the Spin Analytical CFA instrument could handle it all OK?
> 
Arthur,

band sedimentation is also very useful for the new multiwavelength
optics developed by Helmut Coelfen's lab. All of the HPC optimization
methods implemented in UltraScan (2-dimensional spectrum analysis,
genetic algorithm analysis, molecular weight constrained method, and the
new multiwavelength analysis methods) support band sedimentation and
Vinograd cells. You can also mix standard 2 channel and band-forming
cell experiments for global fits over multiple velocity experiments
with different cell types for added hydrodynamic resolution
(see Brookes E, Cao W, Demeler B.  A two-dimensional spectrum
analysis for sedimentation velocity experiments of mixtures with
heterogeneity in molecular weight and shape. Eur Biophys J. 2009, Epub:
http://www.ncbi.nlm.nih.gov/pubmed/19247646?ordinalpos=1&itool=EntrezSystem2.PEntrez.Pubmed.Pubmed_ResultsPanel.Pubmed_DefaultReportPanel.Pubmed_RVDocSum).

For multiwavelength analysis the advantages of band sedimentation (especially
for higher MW species, as you say) is that the spectra for different components
do not overlap and you can separate species not just hydrodynamically, but
also spectrally. The same principle should apply in the SAXS experiments.

Another method that may prove helpful is analytical buoyant density
gradient sedimentation that will separate species based on densities and
provide static peaks that can be analyzed spectrally. We have used this
successfully for analyzing virus packaging, where different stages of
packaged viruses separate because they contain different amounts of DNA,
and hence have significantly different densities. The spectral patterns
of the peaks differ because of different contributions of DNA and protein,
which can be deconvoluted with UltraScan to identify relative amounts.

I am not sure how long SAXS exposures would take in the AUC, but perhaps
having this static peak separated out may be useful, compared to a rapidly
moving boundary.

-Borries
_______________________________________________
RASMB mailing list
RASMB at rasmb.bbri.org
http://rasmb.bbri.org/mailman/listinfo/rasmb