[RASMB] error estimates and F-statistics in Sedphat

Tue Apr 26 19:26:28 PDT 2011

Mark,

The actual data of course looks quite reasonable, however the fit clearly
does not. You also have a significant time-invariant noise component in
these data you should account for. There is another way to fit these
data. In UltraScan (and I believe also in SedAnal) you can fit your
velocity data to an explicit monomer-dimer model, fitting s and D for
both monomer and dimer, as well as the Kd and koff rate constant (and
one or more contaminants, if present, although the more signal goes to
contaminants, the lower the confidence will be in all the other parameters
as well).

For a successful fit it is very helpful to have velocity data for a broad
concentration range as Jack and John already have pointed out. This will
help you get a ball-park estimate for the Kd and allow you to enter some
reasonable constraints for these values in the fitting program. For
example, you do not want to be at too low concentration to where you
essentially get no signal for the dimer, or too concentrated, and not
see any monomer.  Ideally, you want to be somewhere in the middle where
both species are well represented. This can be quickly ascertained with
a van Holde - Weischet analysis of your data.

The complete procedure and an example for how to do it is explained
in this publication:

Demeler B, Brookes E, Wang R, Schirf V, Kim CA. Characterization
of Reversible Associations by Sedimentation Velocity with UltraScan. 
Macromol. Biosci. 2010 Jul 7;10(7):775-82. PMID: 20486142

Good luck, -Borries

> 
> Mark,
>  
> Given that the fit is so bad I'm not sure it makes any sense to try to get a
> confidence interval. The model is clearly not even coming close to
> representing the data, so what does this association constant mean? 
>  
> I agree fully with Jack that you cannot expect the peaks from c(s) analysis
> to give you the monomer and dimer sedimentation coefficients, and that you
> really want to do a dilution series over a big concentration range. For
> absorbance data that usually means either using multiple wavelengths (in
> different runs) or using the 3 mm centerpieces to help cover a big range
> while keeping the OD <=1 for the high end. Then you can either fit the
> weight-average sedimentation coefficient versus concentration to a model, or
> do global whole boundary analysis with SEDPHAT or SEDANAL (with the
> sedimentation coefficients as fitting variables).
>  
> For this or any other experiment I also recommend that you fill you cells
> more completely to get a better separation. Try to get the meniscus at
> 5.9-5.95 cm. 
>  
> John
> 
>   _____  
> 
> From: rasmb-bounces at rasmb.bbri.org [mailto:rasmb-bounces at rasmb.bbri.org] On
> Behalf Of Mark Agacan
> Sent: Tuesday, April 26, 2011 8:36 AM
> To: RASMB at rasmb.bbri.org
> Subject: Re: [RASMB] error estimates and F-statistics in Sedphat
> 
> 
> Thanks Patrick and Chad for the information about error estimates from
> F-statistics.
>  
> My problem now is that the apparent errors for the binding constants derived
> from the analysis of F-statistic appear to be very large.  When I make a
> change to e.g. log10Ka12, by any value from 0.1 - 3.0, there is no change
> (or only a very small change) in chi squared, so I can never reach the
> critical chi squared.  
>  
> Each dataset looks good and gives low rmsd values when processing with the
> c(s) model in sedfit.  I'm using the monomer-dimer self-association model in
> sedphat, with the values for S1 and S2 taken from c(s) analysis.  I have
> tried to fit to other models with only worse results.
>  
> The fit looks very poor (see screenshots attached).  I don't think I need to
> repeat the experiment because the data looks good, at least to my eye, but
> the fit to the monomer-dimer model seems very bad.  
>  
> Any further suggestions or advice would be very welcome.
>  
> Many Thanks,
>  
> Mark
>  
> 
>  
>  
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Dr Mark Agacan, Scientific Officer for the Division of Biological Chemistry
> and Drug Discovery,
> Wellcome Trust Biocentre, College of Life Sciences, University of Dundee,
> Dundee, DD1 5EH 
> Tel: +44 1382 386095    Fax: +44 1382 345764    Mobile: 07525 451 117
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>  
> ************************************************************ 
> Please consider the environment. Do you really need to print this email? 
> 
> 
> >>> Chad Brautigam <Chad.Brautigam at UTSouthwestern.edu> 4/23/2011 15:02 >>>
> 
> Hi, Mark,
> 
> The F-statistics in SEDPHAT are meant to allow you to construct a
> "do-it-yourself" error interval.  Check out the statistics help page.  Under
> the heading "Error Analysis" there is a procedure for how to use the output
> "critical chi-square" from the Statistics Menu to find an error interval.
> 
> Cheers,
> Chad
> 
> 
>   _____  
> 
> From: rasmb-bounces at rasmb.bbri.org [rasmb-bounces at rasmb.bbri.org] on behalf
> of Mark Agacan [M.Agacan at dundee.ac.uk]
> Sent: Friday, April 22, 2011 6:36 AM
> To: rasmb at rasmb-email.bbri.org
> Subject: [RASMB] error estimates and F-statistics in Sedphat
> 
> 
> Hello,
>  
> I'm calculating errors in the binding constants in sedphat as instructed in
> the statistics help page, but I don't appear to get an output for the
> F-statistics.  Are the F-stats output only as (changes in) chi-squared
> values? e.g. the error associated with the binding constants is clearly
> given in the covariance matrix.
>  
> Best Wishes,
>  
> Mark
>  
>  
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Dr Mark Agacan, Scientific Officer for the Division of Biological Chemistry
> and Drug Discovery,
> Wellcome Trust Biocentre, College of Life Sciences, University of Dundee,
> Dundee, DD1 5EH 
> Tel: +44 1382 386095    Fax: +44 1382 345764    Mobile: 07525 451 117
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>  
>  
> ************************************************************ 
> Please consider the environment. Do you really need to print this email? 
> 
> 
> The University of Dundee is a registered Scottish charity, No: SC015096 
> 
> 
>   _____  
> 
> 
> UT Southwestern Medical Center
> The future of medicine, today.
> 
> 
> 
> The University of Dundee is a registered Scottish charity, No: SC015096 
> 
> 
> ------=_NextPart_000_00E0_01CC0417.D24778C0
> Content-Type: text/html;
> 	charset="us-ascii"
> Content-Transfer-Encoding: quoted-printable
> 
> <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
> <HTML dir=3Dltr><HEAD>
> <META content=3D"text/html; charset=3Dus-ascii" =
> http-equiv=3DContent-Type>
> <STYLE id=3DowaParaStyle type=3Dtext/css>
> <!--
> p
> 	{margin-top:0;
> 	margin-bottom:0}
> -->
> </STYLE>
> 
> <META name=3DGENERATOR content=3D"MSHTML 8.00.6001.19048"></HEAD>
> <BODY style=3D"MARGIN: 4px 4px 1px; FONT: 10pt Tahoma">
> <DIV dir=3Dltr align=3Dleft><SPAN =
> class=3D351382420-26042011>Mark,</SPAN></DIV>
> <DIV dir=3Dltr align=3Dleft><SPAN =
> class=3D351382420-26042011></SPAN> </DIV>
> <DIV dir=3Dltr align=3Dleft><SPAN class=3D351382420-26042011>Given that =
> the fit is so=20
> bad I'm not sure it makes any sense to try to get a confidence interval. =
> The=20
> model is clearly not even coming close to representing the data, so what =
> does=20
> this association constant mean? </SPAN></DIV>
> <DIV dir=3Dltr align=3Dleft><SPAN =
> class=3D351382420-26042011></SPAN> </DIV>
> <DIV dir=3Dltr align=3Dleft><SPAN class=3D351382420-26042011>I agree =
> fully with Jack=20
> that you cannot expect the peaks from c(s) analysis to give you the =
> monomer and=20
> dimer sedimentation coefficients, and that you really want to do a =
> dilution=20
> series over a big concentration range. For absorbance data that usually =
> means=20
> either using multiple wavelengths (in different runs) or using the 3 mm=20
> centerpieces to help cover a big range while keeping the OD <=3D1 for =
> the high=20
> end. Then you can either fit the weight-average sedimentation =
> coefficient versus=20
> concentration to a model, or do global whole boundary analysis with =
> SEDPHAT=20
> or SEDANAL (with the sedimentation coefficients as fitting=20
> variables).</SPAN></DIV>
> <DIV dir=3Dltr align=3Dleft><SPAN =
> class=3D351382420-26042011></SPAN> </DIV>
> <DIV dir=3Dltr align=3Dleft><SPAN class=3D351382420-26042011>For this or =
> any other=20
> experiment I also recommend that you fill you cells more completely =
> to get=20
> a better separation. Try to get the meniscus at 5.9-5.95 cm. =
> </SPAN></DIV>
> <DIV dir=3Dltr align=3Dleft><SPAN =
> class=3D351382420-26042011></SPAN> </DIV>
> <DIV dir=3Dltr align=3Dleft><SPAN =
> class=3D351382420-26042011>John</SPAN></DIV><BR>
> <DIV dir=3Dltr lang=3Den-us class=3DOutlookMessageHeader align=3Dleft>
> <HR tabIndex=3D-1>
> <B>From:</B> rasmb-bounces at rasmb.bbri.org =
> [mailto:rasmb-bounces at rasmb.bbri.org]=20
> <B>On Behalf Of </B>Mark Agacan<BR><B>Sent:</B> Tuesday, April 26, 2011 =
> 8:36=20
> AM<BR><B>To:</B> RASMB at rasmb.bbri.org<BR><B>Subject:</B> Re: [RASMB] =
> error=20
> estimates and F-statistics in Sedphat<BR><BR></DIV>
> <DIV></DIV>
> <DIV>Thanks Patrick and Chad for the information about error estimates =
> from=20
> F-statistics.</DIV>
> <DIV> </DIV>
> <DIV>My problem now is that the apparent errors for the binding =
> constants=20
> derived from the analysis of F-statistic appear to be very =
> large. =20
> When I make a change to e.g. log10Ka12, by any value from 0.1 - 3.0, =
> there is no=20
> change (or only a very small change) in chi squared, so I can never =
> reach the=20
> critical chi squared.  </DIV>
> <DIV> </DIV>
> <DIV>Each dataset looks good and gives low rmsd values when processing =
> with the=20
> c(s) model in sedfit.  I'm using the monomer-dimer self-association =
> model=20
> in sedphat, with the values for S1 and S2 taken from c(s) =
> analysis.  I have=20
> tried to fit to other models with only worse results.</DIV>
> <DIV> </DIV>
> <DIV>The fit looks very poor (see screenshots attached).  I don't =
> think I=20
> need to repeat the experiment because the data looks good, at least to =
> my=20
> eye, but the fit to the monomer-dimer model seems very bad.  =
> </DIV>
> <DIV> </DIV>
> <DIV>Any further suggestions or advice would be very welcome.</DIV>
> <DIV> </DIV>
> <DIV>Many Thanks,</DIV>
> <DIV> </DIV>
> <DIV>Mark</DIV>
> <DIV> </DIV>
> <DIV><BR> </DIV>
> <DIV> </DIV>
> <DIV>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~<BR>Dr =
> Mark=20
> Agacan, Scientific Officer for the Division of Biological Chemistry and =
> Drug=20
> Discovery,<BR>Wellcome Trust Biocentre, College of Life Sciences, =
> University of=20
> Dundee, Dundee, DD1 5EH <BR>Tel: +44 1382 386095    Fax: =
> +44 1382=20
> 345764    Mobile: 07525 451=20
> 117<BR>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~</DIV>
> <DIV> </DIV>
> <DIV>
> <DIV>************************************************************ </DIV>
> <DIV>
> <DIV><FONT color=3D#666666>Please consider the environment. Do you =
> really need to=20
> print this email?</FONT> </DIV></DIV><BR><BR>>>> Chad Brautigam =
> 
> <Chad.Brautigam at UTSouthwestern.edu> 4/23/2011 15:02 =
> >>><BR></DIV>
> <DIV=20
> style=3D"FONT-FAMILY: Tahoma; DIRECTION: ltr; COLOR: #000000; FONT-SIZE: =
> 10pt">Hi,=20
> Mark,<BR><BR>The F-statistics in SEDPHAT are meant to allow you to =
> construct a=20
> "do-it-yourself" error interval.  Check out the statistics help =
> page. =20
> Under the heading "Error Analysis" there is a procedure for how to use =
> the=20
> output "critical chi-square" from the Statistics Menu to find an error=20
> interval.<BR><BR>Cheers,<BR>Chad<BR><BR>
> <DIV style=3D"FONT-FAMILY: Times New Roman; COLOR: rgb(0,0,0); =
> FONT-SIZE: 16px">
> <HR tabIndex=3D-1>
> 
> <DIV style=3D"DIRECTION: ltr" id=3DdivRpF398298><FONT color=3D#000000 =
> size=3D2=20
> face=3DTahoma><B>From:</B> rasmb-bounces at rasmb.bbri.org=20
> [rasmb-bounces at rasmb.bbri.org] on behalf of Mark Agacan=20
> [M.Agacan at dundee.ac.uk]<BR><B>Sent:</B> Friday, April 22, 2011 6:36=20
> AM<BR><B>To:</B> rasmb at rasmb-email.bbri.org<BR><B>Subject:</B> [RASMB] =
> error=20
> estimates and F-statistics in Sedphat<BR></FONT><BR></DIV>
> <DIV></DIV>
> <DIV>
> <DIV>Hello,</DIV>
> <DIV> </DIV>
> <DIV>I'm calculating errors in the binding constants in sedphat as =
> instructed in=20
> the statistics help page, but I don't appear to get an output for the=20
> F-statistics.  Are the F-stats output only as (changes in) =
> chi-squared=20
> values? e.g. the error associated with the binding constants is clearly =
> given in=20
> the covariance matrix.</DIV>
> <DIV> </DIV>
> <DIV>Best Wishes,</DIV>
> <DIV> </DIV>
> <DIV>Mark</DIV>
> <DIV> </DIV>
> <DIV> </DIV>
> <DIV>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~<BR>Dr =
> Mark=20
> Agacan, Scientific Officer for the Division of Biological Chemistry and =
> Drug=20
> Discovery,<BR>Wellcome Trust Biocentre, College of Life Sciences, =
> University of=20
> Dundee, Dundee, DD1 5EH <BR>Tel: +44 1382 386095    Fax: =
> +44 1382=20
> 345764    Mobile: 07525 451=20
> 117<BR>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~</DIV>
> <DIV> </DIV>
> <DIV> </DIV>
> <DIV>************************************************************ </DIV>
> <DIV>
> <DIV><FONT color=3D#666666>Please consider the environment. Do you =
> really need to=20
> print this email?</FONT> </DIV></DIV><BR>
> <P>The University of Dundee is a registered Scottish charity, No: =
> SC015096=20
> </P></DIV></DIV></DIV><BR>
> <HR>
> <FONT color=3Dblue size=3D2 face=3DArial><BR>UT Southwestern Medical =
> Center<BR>The=20
> future of medicine, today.<BR></FONT><BR>
> <P>The University of Dundee is a registered Scottish charity, No: =
> SC015096=20
> </P></BODY></HTML>
> 
> ------=_NextPart_000_00E0_01CC0417.D24778C0--
> 
> 
> --===============2060512001==
> Content-Type: text/plain; charset="us-ascii"
> MIME-Version: 1.0
> Content-Transfer-Encoding: 7bit
> Content-Disposition: inline
> 
> _______________________________________________
> RASMB mailing list
> RASMB at rasmb.bbri.org
> http://rasmb.bbri.org/cgi-bin/mailman/listinfo/rasmb
> 
> --===============2060512001==--
> 