On 04/01/2021 10:42, Victor Tobiasson wrote:
Time to join the party, a bit late perhaps but christmas got in the way!
I will mention the speedrun in my presentation tomorrow and hopefully
that may spark a bit of interest. So there are a few days left. I
haven't set a date, maybe 14 Jan.
First of all, great challenge Paul! It was really fun to try to push
my time and I finally got around to implement some scripting which I
put off as well.
My first attempt without prior information (except the
extension sequence and the fact there are three ligands) ~10 min. I
have little to no experience with modelling X-ray data though so it
took a long time to get subsequent runs to regularly meet the
validation metrics. Below is a short outline of my experience so far.
Strategy and approach:
I am using coot via the ccpem 1.5.0 stable release for comparison. I
don't use any more advanced scripts but have several key bindings set
for minimising clicks and mouse micro.
:-) quite necessary.
The general strategy revolves around simply performing the minimal
amount of edits and inputs required to meet the requirements as fast
as possible. I try not to move the center of rotation or the view
manually with the mouse, at some places it's required or faster to do
so but I try to avoid it. The general approach of things to fix are as
follows.
To Do: Fix the rotamers Gln/32, Asp/33 and Phe/89.
That's interesting. Actually 0.17 was quite generous and one need not,
in fact, fix up 32, 33 or 25 to get down there.
Add alternative conformations to Asp/25. Delete the stub and add
Cys/72 with alternative conformations.
72 is deliberately annoying because it is unlike a simple mutation.
Ctrl-D for delete residue is a help here. And then Add Terminal residue
and fast mutate.
Remodel Arg/40 including backbone and rotamers.
Nice, but not needed.
Add terminal residues QTC. Add OXT oxygents to both chains. Add two
SO4 ligands and the 3GP. Add waters with and RMSD of 1.3 with default
distances.
1) Preparations: Open the 'Other modelling tools' panel and load and
place the "accept refinement" popup far away from the main gui, it
gets in the way.
Yes, it does. I don't know why it sometimes resets to centred (I dug
around the code for a day and gave up).
Set the map in the refmac gui.
Interesting comment. I wonder what you mean by it. The refmac gui that I
use allows me to set the model and the data mtz file.
And turn on all restraints in the R/RC menu, set the weight to 200.
200! wow - interesting.
2) Manual edits: Run the script and immediately hit space to center on
A/1. Navigate down the chain to Asp/25 and add an altconf in the most
common rotamer. Further translate until Gln/32-Asp/33, refine and
place the side chains manually as they dont match the rotamers very
well (correct map weight is key here). Accept and continue to Arg/40,
refine and flip the carbonyl. With the right flick the rotamers will
fix themselves. Got to Cys/72, delete it and add a new residue as well
as the alternative conformation. Go to Phe/89 and go for the third
rotamer.
auto-fit rotamer does that for you - I didn't know the rotamer number.
Translate to the end of the chain and add 3 terminals resudies,
serially mutate to QTC.
That's what I do.
Navigate to the end and change the cys rotamer.
For me, the CYS placement routinely interchanges the main chain and the
side chain positions. Interesting that you don't find the same (I presume).
Add OXT for chain A and then B though the other modelling tools panel
(you could script adding OXT to all but felt that was cheating).
If I am allowed to arbitrate on what is cheating, then a function for
"Add OXT to residue 96" is cheating, but "Add OXT to current residue" is
not, "Add OXT to the same residue number as the current residue number
for all chains that have the same sequence" is not cheating either (it
is esoteric, no doubt and not what I did). The different being functions
that are tailored to this structure, as opposed to those that might be
useful for other structures.
Also, I used "NCS Other chain" to get to the B96 (more on that later).
3) Heteroatoms: Open the validate panel and find unmodeled blobs.
It's OK to know that the ligand is at the end of residue A 41.
Go to the first blob and place the 3GP (here I orient the density
before placing as to get the default ligand orientation correct for
the purine).
The orientation of the added ligand, if you add it with "Get Monomer,"
doesn't depend on the orientation of the view.
Quickly refine and fit the phosphate manually.
A speed up here (I think) is, before activating refinement, to select an
orientation of the ligand, where both the phosphate and the base are
misplaced, but, with refinement, they both fall into their respective
densities. I used Ctrl-Arrow keys and Shift Ctrl Arrow keys to do that.
For me, reorienting the ribose was a major part of fitting the ligand.
Interesting that you don't mention it.
Go to the second blob and place SO4. Manually shift the view
"Manually shift the view" I wonder what you mean by that. You can use
"O" ("Other NCS chain") to jump to the same position in the B-chain, add
your SO4 there and jump back (again, with "O" (because there are 2 chains)).
to center on the second SO4 and place it. Click "Find waters..." and
change the threshold to 1.3 (most consistent R/Rfree for me without
manual intervention) and then run.
4) Cleanup: Merge all the ligand chains,
For best speed, I merge before running Refmac.
sphere refine the entire molecule for ~1-2 seconds and run refmac.
(Optional: curse refmac and or yourself for mistakes and time lost)
One thing that this challenge made clear to me is how annoying it was to
have the GUI freeze while running refmac. I will fix that.
The binds and custom settings I use are the following (roughly
following most to least important for this challenge):
e = accept regularization
Return does this (or not?)
!/@/# (SHIFT+1 2 or 3 depending on keyboard layout this changes) =
real space refine zone 3, 7 or 21 residues.
This is a great idea. It's something like the fast selection expansion
in Chimera (but tailored for Coot).
z and x = shift 5 residues along the chain (space * 5)
Ctrl-u 5 M-x space - haha! OK, nice one.
Z and X = shift to the beginning or end of the chain
That is generally useful, in the speed run we can just jump to the problems.
a = add terminal residue
We have "y" in the standard key-bindings set - and that is expanded upon
in the Elaborated Add Terminal Residue in Curlew
r = cycle through rotamers
OK, is that in "intermediate atoms mode" or standard representation? (I
had not thought of the latter).
R = sphere refine (specifically for this challenge I set the radius to
200 to get everything)
We have Chain Refine in Curlew (Shift E). There is no key binding for
"All atom refine" - but there is a menu item.
t = place ligand from library
OK, that would be a speed-up. I use Alt F and then down-arrow 6 times.
q = mutate chain to sequence
And that too? Not sure how you use it - because you'd have to have the
sequence somewhere - or do you mean a residue range?
With this setup and my preliminary route I have tried perhaps a dozen
times seriously and gotten varying results. Initially times would
hover around 3 minutes and then steadily decrease as I made fewer
mistakes. So far my fastest times are;
*Time (fastest accepted): 2 min 29 sec (R/Rfree: 0.170/0.1995) *
*Time (fastest overall): 1 min 42 sec (R/Rfree: 0.169/0.203)*
Let's take the fastest accepted, because the passed the R-free test too.
The times are still very unrefined and I am not really happy with 2.29.
Reviewing your comments, I think that you can go faster.
The big challenge is definitely pushing the Rfree low enough with the
rough edits I make.
Yes, my fastest time is when I did the mods and added the ligand and
only ran Refmac right at the end.
I lose most runs around the 1.50 mark due to Rfree, I never go for
second refinements as it would be too slow
Ah, I see. Makes sense. Similar to my strategy.
and It can definitely be done in one round.
I agree.
I usually lose time when fixing the rotamers for Gln-Asp/32-33,
Now you know that you don't need to. You can get a quicker drop in the
R-factors by adding more waters. Try 1.2 or 1.1.
they are super tricky to do fast and precise.
I think A32 is ligand-related partial occupancy (FWIW).
Also the orienting of the 3GP can be a bother sometimes.
Yes. With the old version of the dictionary, it popped right in, but the
new one, at least in my hands, needs a ribose twist and this is
compounded by the dictionary not allowing JED-Flip on the glycosidic
bond (which I think might be/is an error, but I have not yet discussed
it with my colleagues).
Overall quick clean movements seems to be the way forward. I
believe times around 55 seconds are definitely possible for someone
fast with good mouse movement and macros. I still hesitate several
times and have to think about what to do next. I would be interested
to see if any of you routinely fix some significant mistake I miss!
Hahaha! Yes, I hesitate too "OK, that's done, now what?" I've tried to
memorize 89,41,72,93,63.
Paul
########################################################################
To unsubscribe from the COOT list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=COOT&A=1
This message was issued to members of www.jiscmail.ac.uk/COOT, a mailing list
hosted by www.jiscmail.ac.uk, terms & conditions are available at
https://www.jiscmail.ac.uk/policyandsecurity/
