[Chimera-users] Fitting models into density - meaning of values

[Jessica Holien]

Thanks Tom - so within my map I can say the higher the value the "better" the fit? Luckily we have a fair bit of biological data which adds to the modelling. I have used the authors "suggested" contour level and knew that the correlation should be around 0.8 (or higher) - which it is for these models at this level. This is my first attempt at EM density fitting and thought Chimera was very user friendly!

Jess 

Dr. Jessica Holien, PhD
Research Officer
Biota Structural Biology Laboratory
ACRF Rational Drug Discovery Centre
St. Vincent's Institute of Medical Research

Address: 9 Princes St, Fitzroy, Victoria, Australia 3065
Postal: 41 Victoria St, Fitzroy, Victoria, Australia 3065
Telephone: +61 3 9288 2685
Email: jholien at svi.edu.au

----- Original Message -----
From: "Tom Goddard" <goddard at sonic.net>
To: "Jessica Holien" <jholien at svi.edu.au>
Cc: chimera-users at cgl.ucsf.edu
Sent: Wednesday, 30 October, 2013 12:22:32 PM
Subject: Re: [Chimera-users] Fitting models into density - meaning of values

Hi Jessica,

  The "average map" value reported by Chimera is only useful for comparing different fits in the same map.  The values in a density map normalized in ways that differ from map to map, so the absolute value is meaningless.  What is sometimes reported in the literature is correlation coefficient, ranging from -1 to 1.  The Chimera fitting will report this if you use the Options button on the Fit Map dialog and enable "Use simulated map … with resolution 23".  The correlation compares a simulated map from your atomic model to the experimental map.  Values like 0.8 and higher are often considered "good".  But unfortunately this is a very poor indicator too for several reasons.  First it depends on the region in space over which the correlation was measured.  In Chimera this will be the grid points within the displayed contour of the simulated map.  (You'll have to show the simulated map using the volume dialog or model panel to see that contour level.)  Higher contour levels use fewer grid points and often give a much higher correlation value.  Another reason this is a poor measure is that it doesn't tell you if there are 5 other very different fits with equally good correlation value.  For a low resolution map like 23 Angstroms, there may be many equally good and very different fits.  The short answer is that it is hard to say if a fit to a low resolution map is good in a convincing way, and the research community does not have standard measures for this.  To make your best effort I would try to produce 2nd and 3rd best fits and see if the 1st best one stands out as a visibly better match (and corresponding higher correlation) than the 2nd and 3rd best fits.

        Tom



On Oct 29, 2013, at 5:57 PM, Jessica Holien  wrote:

> I have been using Chimera to fit some homology models into the EM density (23A resolution) of a large complex. Although by eye it looks good I am trying to work out if statistically it is good. Can anyone tell me what is a good "Average Map Value" and what is a bad one? Thanks
> 
> 
> _______________________________________________
> Chimera-users mailing list
> Chimera-users at cgl.ucsf.edu
> http://plato.cgl.ucsf.edu/mailman/listinfo/chimera-users
>