Dear Paul, First and foremost, thanks a lot for your answer.
It is very helpful to get some insight into what's going on with the software, and you explained it simply and exhaustively. I will check the overall quality of the model built in Coot that I used as an input for the last refinement, in order to know if it was good to start with. I hope I will soon find the right strategy! I wish you a nice day Thanks again Silvia Silvia Napolitano ETH Zurich Institute of Molecular Biology and Biophysics Otto-Stern-Weg 5, HPK E14 CH–8093 Zurich Phone: +41 44 6332482 -----Original Message----- From: Paul Emsley <[email protected]> Sent: Thursday, August 6, 2020 6:48 PM To: Napolitano Silvia <[email protected]>; [email protected] Subject: Re: [ccp4bb] Stuck with Refinement On 06/08/2020 16:39, Silvia Napolitano wrote: > Dear CCP4BB community, Dear Silvia, > > I am working on a crystal structure and I am a bit stuck with the refinement. > > To build the model I use Coot and to refine I use Phenix. I may be able to offer some insight about the first part. > > First, I noticed that the geometry analysis of exactly the same files is > quite different between Coot 0.8 and Coot 0.9 (see PowerPoint attached, > slides 1 and 2); isn't that a bit strange? The restraints generation and weighting was completely rewritten between 0.8 and 0.9. I suspect that what you are seeing is a result of using a more sensitive scaling. With some experience, I had noticed that the bars in the geometry analysis as a result of modern refinement were very often too small - in a non-useful way. So I scaled them up a bit. I haven't looked at this sort of side by side analysis for many years so your figures 1 & 2 were a bit of a surprise (that the scaling was so different), but on the whole I think that the current scaling is better. > Anyway, I tried to fix the Ramachandran outliers, geometry, and rotamers, > obtaining, I think, a decent model (see PowerPoint attached, slides 3-5). Yes, they look good. > > However, when I input this model for refinement, the statistics and the > geometry obtained at the end of the refinement cycles are very bad (see > PowerPoint attached, slide 6). A useful data point here is how the same analysis looks at the start of (i.e. before) the refinement. Then one can distinguish between the cases of the refinement being given a good model and chewing it up and the the refinement being given what it thinks is a bad model and throwing up its (metaphorical) hands. Also, the underlying data which is used by Coot is different to that used by Phenix. The Ramachandran plot data is different, the CB distortion model is different, the non-bonded contact model is different - and the means to assess "bad contacts" is different. FWIW, the rotamer database is much the same (or, at least it used to be - I haven't updated it in about 10 years). Regards, Paul. ######################################################################## To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 This message was issued to members of www.jiscmail.ac.uk/CCP4BB, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
