[RASMB] sedfit vs SOMO vs structure

[Thomas Jowitt]

Hi

This has prompted me to do a quick check of the idea of 'random' residuals and the RMSD of a couple of recently run proteins. I have attached a graph of RMSD for a fit (abs 230nm approx 0.7AU, 200 scans fitting every scan) for both proteins vs. successive deletions of the scans. Therefore each point is a fit minus the first, second, third etc. scan. As you can see the fit is appreciably better after discarding the first few scans (nothing new here). The non-random behaviour of the fits in both cases however (protein 1 being a much more well behaved protein) does not diminish until way after the 23rd scan, with a plateau of RMSD occuring somewhere in the 0.00360 region. 

Now, this obviously is not an exhaustive study (just did it in the last 10 mins), but interesting non the less. I normaly leave out the first few scans, but this may have prompted me to look at leaving out rather a few more.

Cheers

Tom

Dr T A Jowitt
Biomolecular Analysis Core Facility Manager
Faculty of Life Sciences, University Of Manchester
Michael Smith Building
Manchester, M13 9PT
Tel; +44 (0)161 3065176

________________________________________
From: rasmb-bounces at rasmb.bbri.org [rasmb-bounces at rasmb.bbri.org] on behalf of Borries Demeler [demeler at biochem.uthscsa.edu]
Sent: 11 July 2011 15:16
To: Chad Brautigam
Cc: rasmb at rasmb.bbri.org
Subject: Re: [RASMB] sedfit vs SOMO vs structure

>
> Hi, Andrew,
>
> I agree that Borries' number seems a bit low.  0.004 to 0.005 seems to be a good number on our machine.
>
> I suspect that Borries keeps his lamps and slits very clean.  I also suspect that the number he is quoting is from pseduoabsorbance measurements, but I could be wrong.  Care to comment, Borries?
>
> Cheers,
> Chad

Yes, Chad, you are absolutely correct! Pseudo-absorbance will give you
a factor of ~ 1.4 improvement in stochastic noise. And yes, our Beckman
service does a phantastic job to keep our 15 year old XLAs operating
optimally. The rest is achieved by regular cleaning of the lamp.

I want to emphasize that reducing noise is just as important as fitting
with the correct model in order to derive meaningful information from
any data. If you have non-random residuals all bets are off.

I also want to mention that the appearance of "two minima" is nothing
unusual at all, it really is a reflection of a fundamental principle:

When you fit noisy data there is no guarantee of a UNIQUE solution,
in fact there may be an infinite number of solutions that will all give
the same RMSD. The principle also suggests that a reduction in random
noise also diminishes this problem.  The best way around this problem is
a Monte Carlo analysis of your model to map out the confidence regions
of the solution, which tells you what you really want to know: what is
the chance that my model is correct?

RMSD values from absorbance data around 0.009 when fitted with a finite
element solution suggests there is something wrong with the model or
the instrument, or both.

Best wishes, -Borries
_______________________________________________
RASMB mailing list
RASMB at rasmb.bbri.org
http://rasmb.bbri.org/cgi-bin/mailman/listinfo/rasmb
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Untitled.jpg
Type: image/jpeg
Size: 56245 bytes
Desc: Untitled.jpg
URL: <http://list.rasmb.org/pipermail/rasmb-rasmb.org/attachments/20110711/ff219993/attachment.jpg>