Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrin and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin (https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
* first, check carefully for entanglement between distant domains (this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem. * Start ISOLDE. Then, assuming your protein is model #1, do the following commands: * isolde restrain distances #1 distanceCutoff 5 * isolde restrain torsions #1 angleRange 60 * isolde sim start #1 * Then, you'll want to use the "tug selection" mouse mode (third from bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle. * If you have unwanted distance restraints between domains that should be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel". * You might find it easier for a job like this to reduce the view to a C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
________________________________ From: ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edu Sent: 23 October 2021 16:30 To: chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu Subject: [chimerax-users] Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrin and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
________________________________ This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
Hi Roden, Tristan,
I tried Tristan's suggestion for straightening the long AlphaFold helix model. It didn't work well for me because the restraints were too soft, so if I pulled on the first 160 residues it hardly moved the attached 2000 and instead just distorted the linker. Tristan is there a way to make the restraints much stiffer? I see the "isolde restrain distance" command has a "kappa" option and "falloff" option but I don't know if kappa is a stiffness or what magnitude it usually uses.
I think a better approach is to just rigidly rotate and translate large straight sections by hand using the "Move atoms" mouse mode in the ChimeraX 1.3 toolbar tab called Right Mouse. I tried this and have attached an image of about 10 minutes fiddling by hand. I hover the mouse at a hinge residue to get the residue number and then select everything on one side of that hinge "select :1182-2419" then use the right mouse button in the "move atoms" mode, holding the shift key down to do rotations, and no shift key to do translations. I zoom in on the hinge to try to adjust so the bond is not stretched and to avoid clashes. It would probably take 1 hour to do a good job. With my 10 minute test I then saved an mmCIF file and tried in ChimeraX 1.2.5 to minimize it with ISOLDE, but it said there were severe clashes.
Tom
On 10/24/2021 1:49 AM, Tristan Croll via ChimeraX-users wrote:
Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin (https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
- first, check carefully for entanglement between distant domains (this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem.
- Start ISOLDE. Then, assuming your protein is model #1, do the following commands: o isolde restrain distances #1 distanceCutoff 5 o isolde restrain torsions #1 angleRange 60 o isolde sim start #1
- Then, you'll want to use the "tug selection" mouse mode (third from bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle.
- If you have unwanted distance restraints between domains that should be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel".
- You might find it easier for a job like this to reduce the view to a C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
*From:* ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edu *Sent:* 23 October 2021 16:30 *To:* chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu *Subject:* [chimerax-users] Refine predicted structure by pulling Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549 https://alphafold.ebi.ac.uk/entry/P02549. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrin https://en.wikipedia.org/wiki/Spectrin and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
Hi Tristan and Tom,
Thanks very much! Very appreciated! I did play with ISOLDE with the mentioned guidance. I faced the same issues as mentioned by Tom. I thought I should try it a few more times and then reply.
So, I encountered this "at least one residue in your model does not match any templates in the MD forcefiled" warning first; and then it asked me to "addh". And then the same as Tom mentioned. The force I have is more of a "directional" force. I wonder is it possible to have a torque force? Or, is it possible to set a group of amino acids as one rigid body?
Very appreciate the fiddling (manual modeling) guidance too.
Roden
On Mon, Oct 25, 2021 at 10:02 PM Tom Goddard goddard@sonic.net wrote:
Hi Roden, Tristan,
I tried Tristan's suggestion for straightening the long AlphaFold helix model. It didn't work well for me because the restraints were too soft, so if I pulled on the first 160 residues it hardly moved the attached 2000 and instead just distorted the linker. Tristan is there a way to make the restraints much stiffer? I see the "isolde restrain distance" command has a "kappa" option and "falloff" option but I don't know if kappa is a stiffness or what magnitude it usually uses.
I think a better approach is to just rigidly rotate and translate large straight sections by hand using the "Move atoms" mouse mode in the ChimeraX 1.3 toolbar tab called Right Mouse. I tried this and have attached an image of about 10 minutes fiddling by hand. I hover the mouse at a hinge residue to get the residue number and then select everything on one side of that hinge "select :1182-2419" then use the right mouse button in the "move atoms" mode, holding the shift key down to do rotations, and no shift key to do translations. I zoom in on the hinge to try to adjust so the bond is not stretched and to avoid clashes. It would probably take 1 hour to do a good job. With my 10 minute test I then saved an mmCIF file and tried in ChimeraX 1.2.5 to minimize it with ISOLDE, but it said there were severe clashes.
TomOn 10/24/2021 1:49 AM, Tristan Croll via ChimeraX-users wrote:
Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin ( https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde https://urldefense.com/v3/__https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde__;!!Nmw4Hv0!jtJ8dNnTQ0tJc0CLw8HqKh7NIbLUHH72Nc6AJZpi2582f8VOx0v-FbJqs87FyKfCyVw$). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
- first, check carefully for entanglement between distant domains
(this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem.
- Start ISOLDE. Then, assuming your protein is model #1, do the
following commands: - isolde restrain distances #1 distanceCutoff 5 - isolde restrain torsions #1 angleRange 60 - isolde sim start #1
- Then, you'll want to use the "tug selection" mouse mode (third from
bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle.
- If you have unwanted distance restraints between domains that should
be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel".
- You might find it easier for a job like this to reduce the view to a
C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
*From:* ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu *Sent:* 23 October 2021 16:30 *To:* chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu *Subject:* [chimerax-users] Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549 https://urldefense.com/v3/__https://alphafold.ebi.ac.uk/entry/P02549__;!!Nmw4Hv0!jtJ8dNnTQ0tJc0CLw8HqKh7NIbLUHH72Nc6AJZpi2582f8VOx0v-FbJqs87FZ78SQ3E$. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrin https://urldefense.com/v3/__https://en.wikipedia.org/wiki/Spectrin__;!!Nmw4Hv0!jtJ8dNnTQ0tJc0CLw8HqKh7NIbLUHH72Nc6AJZpi2582f8VOx0v-FbJqs87F7tqKeq8$ and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/ https://urldefense.com/v3/__https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/__;!!Nmw4Hv0!jtJ8dNnTQ0tJc0CLw8HqKh7NIbLUHH72Nc6AJZpi2582f8VOx0v-FbJqs87FiXfOrTA$. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
ChimeraX-users mailing listChimeraX-users@cgl.ucsf.edu Manage subscription:https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!jtJ8dNnTQ0tJc0CLw8HqKh7NIbLUHH72Nc6AJZpi2582f8VOx0v-FbJqs87F2XryFwk$
Hi Tom,
Yes, "kappa" is the stiffness. Default in the current release is 5, (as specified in the documentation via "usage isolde restrain") which I agree is very soft for this particular purpose - after playing around with it a bit I find "isolde restrain distances #1 distance 5 kappa 100 fallOff -3" works quite well. The extra "adjustForConfidence true" argument coming in the new ISOLDE release (not yet available - here in a few weeks) will help further - because it uses AlphaFold's own confidence on residue-residue distances, it can extend the distance restraints over a longer range (e.g. the default 8 Angstroms) while ignoring spurious chance contacts between residues. Will play around a little more and share the result.
-- Tristan ________________________________ From: Tom Goddard goddard@sonic.net Sent: 25 October 2021 20:01 To: chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu; Roden Deng Luo deng.luo@kaust.edu.sa; Tristan Croll tic20@cam.ac.uk Subject: Re: [chimerax-users] Refine predicted structure by pulling
Hi Roden, Tristan,
I tried Tristan's suggestion for straightening the long AlphaFold helix model. It didn't work well for me because the restraints were too soft, so if I pulled on the first 160 residues it hardly moved the attached 2000 and instead just distorted the linker. Tristan is there a way to make the restraints much stiffer? I see the "isolde restrain distance" command has a "kappa" option and "falloff" option but I don't know if kappa is a stiffness or what magnitude it usually uses.
I think a better approach is to just rigidly rotate and translate large straight sections by hand using the "Move atoms" mouse mode in the ChimeraX 1.3 toolbar tab called Right Mouse. I tried this and have attached an image of about 10 minutes fiddling by hand. I hover the mouse at a hinge residue to get the residue number and then select everything on one side of that hinge "select :1182-2419" then use the right mouse button in the "move atoms" mode, holding the shift key down to do rotations, and no shift key to do translations. I zoom in on the hinge to try to adjust so the bond is not stretched and to avoid clashes. It would probably take 1 hour to do a good job. With my 10 minute test I then saved an mmCIF file and tried in ChimeraX 1.2.5 to minimize it with ISOLDE, but it said there were severe clashes.
Tom
On 10/24/2021 1:49 AM, Tristan Croll via ChimeraX-users wrote: Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin (https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
* first, check carefully for entanglement between distant domains (this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem. * Start ISOLDE. Then, assuming your protein is model #1, do the following commands: * isolde restrain distances #1 distanceCutoff 5 * isolde restrain torsions #1 angleRange 60 * isolde sim start #1 * Then, you'll want to use the "tug selection" mouse mode (third from bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle. * If you have unwanted distance restraints between domains that should be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel". * You might find it easier for a job like this to reduce the view to a C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
________________________________ From: ChimeraX-users chimerax-users-bounces@cgl.ucsf.edumailto:chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu Sent: 23 October 2021 16:30 To: chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu Subject: [chimerax-users] Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrin and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
________________________________ This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edumailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
Result of some quite aggressive stretching (alternating pulling on residues 2000 onwards and 1-700, extending the tugged selection at both ends as things straightened out). This was with my dev ISOLDE build allowing me to use the "adjustForConfidence true" argument and an 8 Angstrom distance cutoff, which definitely helped keep things under control (also ended up restraining torsions with "isolde restrain torsions #1". Quick-and-dirty video overview of the final model at https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=s..., showing that almost all the distance restraints remain well satisfied, except for a few at the hinges.
Hope you find it useful!
Tristan
________________________________ From: ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu on behalf of Tristan Croll via ChimeraX-users chimerax-users@cgl.ucsf.edu Sent: 25 October 2021 21:14 To: goddard goddard@sonic.net; chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu; Roden Deng Luo deng.luo@kaust.edu.sa Subject: Re: [chimerax-users] Refine predicted structure by pulling
Hi Tom,
Yes, "kappa" is the stiffness. Default in the current release is 5, (as specified in the documentation via "usage isolde restrain") which I agree is very soft for this particular purpose - after playing around with it a bit I find "isolde restrain distances #1 distance 5 kappa 100 fallOff -3" works quite well. The extra "adjustForConfidence true" argument coming in the new ISOLDE release (not yet available - here in a few weeks) will help further - because it uses AlphaFold's own confidence on residue-residue distances, it can extend the distance restraints over a longer range (e.g. the default 8 Angstroms) while ignoring spurious chance contacts between residues. Will play around a little more and share the result.
-- Tristan ________________________________ From: Tom Goddard goddard@sonic.net Sent: 25 October 2021 20:01 To: chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu; Roden Deng Luo deng.luo@kaust.edu.sa; Tristan Croll tic20@cam.ac.uk Subject: Re: [chimerax-users] Refine predicted structure by pulling
Hi Roden, Tristan,
I tried Tristan's suggestion for straightening the long AlphaFold helix model. It didn't work well for me because the restraints were too soft, so if I pulled on the first 160 residues it hardly moved the attached 2000 and instead just distorted the linker. Tristan is there a way to make the restraints much stiffer? I see the "isolde restrain distance" command has a "kappa" option and "falloff" option but I don't know if kappa is a stiffness or what magnitude it usually uses.
I think a better approach is to just rigidly rotate and translate large straight sections by hand using the "Move atoms" mouse mode in the ChimeraX 1.3 toolbar tab called Right Mouse. I tried this and have attached an image of about 10 minutes fiddling by hand. I hover the mouse at a hinge residue to get the residue number and then select everything on one side of that hinge "select :1182-2419" then use the right mouse button in the "move atoms" mode, holding the shift key down to do rotations, and no shift key to do translations. I zoom in on the hinge to try to adjust so the bond is not stretched and to avoid clashes. It would probably take 1 hour to do a good job. With my 10 minute test I then saved an mmCIF file and tried in ChimeraX 1.2.5 to minimize it with ISOLDE, but it said there were severe clashes.
Tom
On 10/24/2021 1:49 AM, Tristan Croll via ChimeraX-users wrote: Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin (https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
* first, check carefully for entanglement between distant domains (this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem. * Start ISOLDE. Then, assuming your protein is model #1, do the following commands: * isolde restrain distances #1 distanceCutoff 5 * isolde restrain torsions #1 angleRange 60 * isolde sim start #1 * Then, you'll want to use the "tug selection" mouse mode (third from bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle. * If you have unwanted distance restraints between domains that should be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel". * You might find it easier for a job like this to reduce the view to a C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
________________________________ From: ChimeraX-users chimerax-users-bounces@cgl.ucsf.edumailto:chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu Sent: 23 October 2021 16:30 To: chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu Subject: [chimerax-users] Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrin and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
________________________________ This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edumailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
Dear Tristan,
Thanks! I think this is exactly what I have been looking forward to for quite some time. A few follow-up questions.
1. What is your hardware and how long does it take to reach the state you showed? I got one RTX 3090 on my local workstation (dual boot for Win10 and Ubuntu). Additional four A6000 on a remote ubuntu server and V100s on a cluster. Do you think I would need to bother with setting up the workflow and X11 forwarding on the remote machines or one RTX 3090 is fine for daily tasks? 2. I feel intimidated by the jargons mentioned above. I found this documentation (https://isolde.cimr.cam.ac.uk/documentation/). Apart from that, do you have other suggestions for reading materials? 3. Would it be possible to have multiple proteins in the scene and incorporate chemical crosslink data? It gives us the information of amino acid X should be within e.g. 30 Å to amino acid Y. 4. Would it be possible to constrain the structure inside a general mesh (e.g. obj file) rather than an EM density map?
Thanks, Roden
On Mon, Oct 25, 2021 at 11:37 PM Tristan Croll tic20@cam.ac.uk wrote:
Result of some quite aggressive stretching (alternating pulling on residues 2000 onwards and 1-700, extending the tugged selection at both ends as things straightened out). This was with my dev ISOLDE build allowing me to use the "adjustForConfidence true" argument and an 8 Angstrom distance cutoff, which definitely helped keep things under control (also ended up restraining torsions with "isolde restrain torsions #1". Quick-and-dirty video overview of the final model at https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=s... https://urldefense.com/v3/__https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=sharing__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSS_JfJXg$, showing that almost all the distance restraints remain well satisfied, except for a few at the hinges.
Hope you find it useful!
Tristan
*From:* ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu on behalf of Tristan Croll via ChimeraX-users chimerax-users@cgl.ucsf.edu *Sent:* 25 October 2021 21:14 *To:* goddard goddard@sonic.net; chimerax-users@cgl.ucsf.edu < chimerax-users@cgl.ucsf.edu>; Roden Deng Luo deng.luo@kaust.edu.sa *Subject:* Re: [chimerax-users] Refine predicted structure by pulling
Hi Tom,
Yes, "kappa" is the stiffness. Default in the current release is 5, (as specified in the documentation via "usage isolde restrain") which I agree is very soft for this particular purpose - after playing around with it a bit I find "isolde restrain distances #1 distance 5 kappa 100 fallOff -3" works quite well. The extra "adjustForConfidence true" argument coming in the new ISOLDE release (not yet available - here in a few weeks) will help further - because it uses AlphaFold's own confidence on residue-residue distances, it can extend the distance restraints over a longer range (e.g. the default 8 Angstroms) while ignoring spurious chance contacts between residues. Will play around a little more and share the result.
-- Tristan
*From:* Tom Goddard goddard@sonic.net *Sent:* 25 October 2021 20:01 *To:* chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu; Roden Deng Luo deng.luo@kaust.edu.sa; Tristan Croll tic20@cam.ac.uk *Subject:* Re: [chimerax-users] Refine predicted structure by pulling
Hi Roden, Tristan,
I tried Tristan's suggestion for straightening the long AlphaFold helix model. It didn't work well for me because the restraints were too soft, so if I pulled on the first 160 residues it hardly moved the attached 2000 and instead just distorted the linker. Tristan is there a way to make the restraints much stiffer? I see the "isolde restrain distance" command has a "kappa" option and "falloff" option but I don't know if kappa is a stiffness or what magnitude it usually uses.
I think a better approach is to just rigidly rotate and translate large straight sections by hand using the "Move atoms" mouse mode in the ChimeraX 1.3 toolbar tab called Right Mouse. I tried this and have attached an image of about 10 minutes fiddling by hand. I hover the mouse at a hinge residue to get the residue number and then select everything on one side of that hinge "select :1182-2419" then use the right mouse button in the "move atoms" mode, holding the shift key down to do rotations, and no shift key to do translations. I zoom in on the hinge to try to adjust so the bond is not stretched and to avoid clashes. It would probably take 1 hour to do a good job. With my 10 minute test I then saved an mmCIF file and tried in ChimeraX 1.2.5 to minimize it with ISOLDE, but it said there were severe clashes.
TomOn 10/24/2021 1:49 AM, Tristan Croll via ChimeraX-users wrote:
Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin ( https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde https://urldefense.com/v3/__https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS07ItNcc$). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
- first, check carefully for entanglement between distant domains
(this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem.
- Start ISOLDE. Then, assuming your protein is model #1, do the
following commands: - isolde restrain distances #1 distanceCutoff 5 - isolde restrain torsions #1 angleRange 60 - isolde sim start #1
- Then, you'll want to use the "tug selection" mouse mode (third from
bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle.
- If you have unwanted distance restraints between domains that should
be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel".
- You might find it easier for a job like this to reduce the view to a
C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
*From:* ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu *Sent:* 23 October 2021 16:30 *To:* chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu *Subject:* [chimerax-users] Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549 https://urldefense.com/v3/__https://alphafold.ebi.ac.uk/entry/P02549__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS138lyq8$. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrin https://urldefense.com/v3/__https://en.wikipedia.org/wiki/Spectrin__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSPVkUGws$ and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/ https://urldefense.com/v3/__https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSkI3UiIc$. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
ChimeraX-users mailing listChimeraX-users@cgl.ucsf.edu Manage subscription:https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSxjHBVGk$
Hi Roden,
Glad you're happy!
1. ISOLDE is currently only capable of using a single GPU, so your big cluster isn't going to help much here. This was done on a laptop with a RTX 2080 GPU, and took about 20-30 minutes. 2. Try "usage isolde restrain", click one of the links that appears in the log, and you'll get the most up-to-date documentation for the command. I know these particular commands are a little on the complex side and I am trying to work out ways to make them more user-friendly, but persevere and you'll get the hang of it. Don't be afraid to experiment (just save your work first!). 3. In theory incorporation of chemical crosslinking data is possible, but right now I'm afraid it's only available through the Python layer (i.e. you'll need to do a little scripting). One of many things I'm hoping to flesh out in future. 4. There's no functionality at present to constrain within a mesh with hard boundaries, since everything in ISOLDE is gradient driven. You'd need to figure out a way to convert your mesh into some form of simulated map (i.e. with gradually increasing "density" values from the surface to the centre).
Best regards, Tristan
________________________________ From: Roden Deng Luo deng.luo@kaust.edu.sa Sent: 26 October 2021 12:00 To: Tristan Croll tic20@cam.ac.uk Cc: goddard goddard@sonic.net; chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu Subject: Re: [chimerax-users] Refine predicted structure by pulling
Dear Tristan,
Thanks! I think this is exactly what I have been looking forward to for quite some time. A few follow-up questions.
1. What is your hardware and how long does it take to reach the state you showed? I got one RTX 3090 on my local workstation (dual boot for Win10 and Ubuntu). Additional four A6000 on a remote ubuntu server and V100s on a cluster. Do you think I would need to bother with setting up the workflow and X11 forwarding on the remote machines or one RTX 3090 is fine for daily tasks? 2. I feel intimidated by the jargons mentioned above. I found this documentation (https://isolde.cimr.cam.ac.uk/documentation/). Apart from that, do you have other suggestions for reading materials? 3. Would it be possible to have multiple proteins in the scene and incorporate chemical crosslink data? It gives us the information of amino acid X should be within e.g. 30 Å to amino acid Y. 4. Would it be possible to constrain the structure inside a general mesh (e.g. obj file) rather than an EM density map?
Thanks, Roden
On Mon, Oct 25, 2021 at 11:37 PM Tristan Croll <tic20@cam.ac.ukmailto:tic20@cam.ac.uk> wrote: Result of some quite aggressive stretching (alternating pulling on residues 2000 onwards and 1-700, extending the tugged selection at both ends as things straightened out). This was with my dev ISOLDE build allowing me to use the "adjustForConfidence true" argument and an 8 Angstrom distance cutoff, which definitely helped keep things under control (also ended up restraining torsions with "isolde restrain torsions #1". Quick-and-dirty video overview of the final model at https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=s...https://urldefense.com/v3/__https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=sharing__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSS_JfJXg$, showing that almost all the distance restraints remain well satisfied, except for a few at the hinges.
Hope you find it useful!
Tristan
________________________________ From: ChimeraX-users <chimerax-users-bounces@cgl.ucsf.edumailto:chimerax-users-bounces@cgl.ucsf.edu> on behalf of Tristan Croll via ChimeraX-users <chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu> Sent: 25 October 2021 21:14 To: goddard <goddard@sonic.netmailto:goddard@sonic.net>; chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu <chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu>; Roden Deng Luo <deng.luo@kaust.edu.samailto:deng.luo@kaust.edu.sa> Subject: Re: [chimerax-users] Refine predicted structure by pulling
Hi Tom,
Yes, "kappa" is the stiffness. Default in the current release is 5, (as specified in the documentation via "usage isolde restrain") which I agree is very soft for this particular purpose - after playing around with it a bit I find "isolde restrain distances #1 distance 5 kappa 100 fallOff -3" works quite well. The extra "adjustForConfidence true" argument coming in the new ISOLDE release (not yet available - here in a few weeks) will help further - because it uses AlphaFold's own confidence on residue-residue distances, it can extend the distance restraints over a longer range (e.g. the default 8 Angstroms) while ignoring spurious chance contacts between residues. Will play around a little more and share the result.
-- Tristan ________________________________ From: Tom Goddard <goddard@sonic.netmailto:goddard@sonic.net> Sent: 25 October 2021 20:01 To: chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu <chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu>; Roden Deng Luo <deng.luo@kaust.edu.samailto:deng.luo@kaust.edu.sa>; Tristan Croll <tic20@cam.ac.ukmailto:tic20@cam.ac.uk> Subject: Re: [chimerax-users] Refine predicted structure by pulling
Hi Roden, Tristan,
I tried Tristan's suggestion for straightening the long AlphaFold helix model. It didn't work well for me because the restraints were too soft, so if I pulled on the first 160 residues it hardly moved the attached 2000 and instead just distorted the linker. Tristan is there a way to make the restraints much stiffer? I see the "isolde restrain distance" command has a "kappa" option and "falloff" option but I don't know if kappa is a stiffness or what magnitude it usually uses.
I think a better approach is to just rigidly rotate and translate large straight sections by hand using the "Move atoms" mouse mode in the ChimeraX 1.3 toolbar tab called Right Mouse. I tried this and have attached an image of about 10 minutes fiddling by hand. I hover the mouse at a hinge residue to get the residue number and then select everything on one side of that hinge "select :1182-2419" then use the right mouse button in the "move atoms" mode, holding the shift key down to do rotations, and no shift key to do translations. I zoom in on the hinge to try to adjust so the bond is not stretched and to avoid clashes. It would probably take 1 hour to do a good job. With my 10 minute test I then saved an mmCIF file and tried in ChimeraX 1.2.5 to minimize it with ISOLDE, but it said there were severe clashes.
Tom
On 10/24/2021 1:49 AM, Tristan Croll via ChimeraX-users wrote: Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin (https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisoldehttps://urldefense.com/v3/__https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS07ItNcc$). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
* first, check carefully for entanglement between distant domains (this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem. * Start ISOLDE. Then, assuming your protein is model #1, do the following commands: * isolde restrain distances #1 distanceCutoff 5 * isolde restrain torsions #1 angleRange 60 * isolde sim start #1 * Then, you'll want to use the "tug selection" mouse mode (third from bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle. * If you have unwanted distance restraints between domains that should be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel". * You might find it easier for a job like this to reduce the view to a C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
________________________________ From: ChimeraX-users chimerax-users-bounces@cgl.ucsf.edumailto:chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu Sent: 23 October 2021 16:30 To: chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu Subject: [chimerax-users] Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549https://urldefense.com/v3/__https://alphafold.ebi.ac.uk/entry/P02549__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS138lyq8$. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrinhttps://urldefense.com/v3/__https://en.wikipedia.org/wiki/Spectrin__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSPVkUGws$ and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/https://urldefense.com/v3/__https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSkI3UiIc$. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
________________________________ This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edumailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-usershttps://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSxjHBVGk$
________________________________ This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
Hi Tristan, Many thanks! I will get myself prepared for ISOLDE. If it is indeed possible to add in crosslinking and possible for multiple proteins, would it be fine for you to write a short description of how to incorporate the crosslinking data at the Python layer (not urgent at all, just at your convenience)? (It has been a while since I deviated from Python and dived into c++ and CUDA for graphics and parallel computing. But it should be fine with a quick review on Python syntax and then a few examples of achieving crosslinking between two demo structures).
Best, Roden
On Tue, Oct 26, 2021 at 2:10 PM Tristan Croll tic20@cam.ac.uk wrote:
Hi Roden,
Glad you're happy!
- ISOLDE is currently only capable of using a single GPU, so your big
cluster isn't going to help much here. This was done on a laptop with a RTX 2080 GPU, and took about 20-30 minutes. 2. Try "usage isolde restrain", click one of the links that appears in the log, and you'll get the most up-to-date documentation for the command. I know these particular commands are a little on the complex side and I am trying to work out ways to make them more user-friendly, but persevere and you'll get the hang of it. Don't be afraid to experiment (just save your work first!). 3. In theory incorporation of chemical crosslinking data is possible, but right now I'm afraid it's only available through the Python layer (i.e. you'll need to do a little scripting). One of many things I'm hoping to flesh out in future. 4. There's no functionality at present to constrain within a mesh with hard boundaries, since everything in ISOLDE is gradient driven. You'd need to figure out a way to convert your mesh into some form of simulated map (i.e. with gradually increasing "density" values from the surface to the centre).
Best regards, Tristan
*From:* Roden Deng Luo deng.luo@kaust.edu.sa *Sent:* 26 October 2021 12:00 *To:* Tristan Croll tic20@cam.ac.uk *Cc:* goddard goddard@sonic.net; chimerax-users@cgl.ucsf.edu < chimerax-users@cgl.ucsf.edu> *Subject:* Re: [chimerax-users] Refine predicted structure by pulling
Dear Tristan,
Thanks! I think this is exactly what I have been looking forward to for quite some time. A few follow-up questions.
- What is your hardware and how long does it take to reach the state you
showed? I got one RTX 3090 on my local workstation (dual boot for Win10 and Ubuntu). Additional four A6000 on a remote ubuntu server and V100s on a cluster. Do you think I would need to bother with setting up the workflow and X11 forwarding on the remote machines or one RTX 3090 is fine for daily tasks? 2. I feel intimidated by the jargons mentioned above. I found this documentation (https://isolde.cimr.cam.ac.uk/documentation/ https://urldefense.com/v3/__https://isolde.cimr.cam.ac.uk/documentation/__;!!Nmw4Hv0!lzCpbVPCotIxWeqmSBW4kHWs5yl7AFeWsveYGcv3E0N6fpto92B_Byoio2SdfQcFrdM$). Apart from that, do you have other suggestions for reading materials? 3. Would it be possible to have multiple proteins in the scene and incorporate chemical crosslink data? It gives us the information of amino acid X should be within e.g. 30 Å to amino acid Y. 4. Would it be possible to constrain the structure inside a general mesh (e.g. obj file) rather than an EM density map?
Thanks, Roden
On Mon, Oct 25, 2021 at 11:37 PM Tristan Croll tic20@cam.ac.uk wrote:
Result of some quite aggressive stretching (alternating pulling on residues 2000 onwards and 1-700, extending the tugged selection at both ends as things straightened out). This was with my dev ISOLDE build allowing me to use the "adjustForConfidence true" argument and an 8 Angstrom distance cutoff, which definitely helped keep things under control (also ended up restraining torsions with "isolde restrain torsions #1". Quick-and-dirty video overview of the final model at https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=s... https://urldefense.com/v3/__https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=sharing__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSS_JfJXg$, showing that almost all the distance restraints remain well satisfied, except for a few at the hinges.
Hope you find it useful!
Tristan
*From:* ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu on behalf of Tristan Croll via ChimeraX-users chimerax-users@cgl.ucsf.edu *Sent:* 25 October 2021 21:14 *To:* goddard goddard@sonic.net; chimerax-users@cgl.ucsf.edu < chimerax-users@cgl.ucsf.edu>; Roden Deng Luo deng.luo@kaust.edu.sa *Subject:* Re: [chimerax-users] Refine predicted structure by pulling
Hi Tom,
Yes, "kappa" is the stiffness. Default in the current release is 5, (as specified in the documentation via "usage isolde restrain") which I agree is very soft for this particular purpose - after playing around with it a bit I find "isolde restrain distances #1 distance 5 kappa 100 fallOff -3" works quite well. The extra "adjustForConfidence true" argument coming in the new ISOLDE release (not yet available - here in a few weeks) will help further - because it uses AlphaFold's own confidence on residue-residue distances, it can extend the distance restraints over a longer range (e.g. the default 8 Angstroms) while ignoring spurious chance contacts between residues. Will play around a little more and share the result.
-- Tristan
*From:* Tom Goddard goddard@sonic.net *Sent:* 25 October 2021 20:01 *To:* chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu; Roden Deng Luo deng.luo@kaust.edu.sa; Tristan Croll tic20@cam.ac.uk *Subject:* Re: [chimerax-users] Refine predicted structure by pulling
Hi Roden, Tristan,
I tried Tristan's suggestion for straightening the long AlphaFold helix model. It didn't work well for me because the restraints were too soft, so if I pulled on the first 160 residues it hardly moved the attached 2000 and instead just distorted the linker. Tristan is there a way to make the restraints much stiffer? I see the "isolde restrain distance" command has a "kappa" option and "falloff" option but I don't know if kappa is a stiffness or what magnitude it usually uses.
I think a better approach is to just rigidly rotate and translate large straight sections by hand using the "Move atoms" mouse mode in the ChimeraX 1.3 toolbar tab called Right Mouse. I tried this and have attached an image of about 10 minutes fiddling by hand. I hover the mouse at a hinge residue to get the residue number and then select everything on one side of that hinge "select :1182-2419" then use the right mouse button in the "move atoms" mode, holding the shift key down to do rotations, and no shift key to do translations. I zoom in on the hinge to try to adjust so the bond is not stretched and to avoid clashes. It would probably take 1 hour to do a good job. With my 10 minute test I then saved an mmCIF file and tried in ChimeraX 1.2.5 to minimize it with ISOLDE, but it said there were severe clashes.
TomOn 10/24/2021 1:49 AM, Tristan Croll via ChimeraX-users wrote:
Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin ( https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde https://urldefense.com/v3/__https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS07ItNcc$). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
- first, check carefully for entanglement between distant domains
(this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem.
- Start ISOLDE. Then, assuming your protein is model #1, do the
following commands: - isolde restrain distances #1 distanceCutoff 5 - isolde restrain torsions #1 angleRange 60 - isolde sim start #1
- Then, you'll want to use the "tug selection" mouse mode (third from
bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle.
- If you have unwanted distance restraints between domains that should
be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel".
- You might find it easier for a job like this to reduce the view to a
C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
*From:* ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu *Sent:* 23 October 2021 16:30 *To:* chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu *Subject:* [chimerax-users] Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549 https://urldefense.com/v3/__https://alphafold.ebi.ac.uk/entry/P02549__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS138lyq8$. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrin https://urldefense.com/v3/__https://en.wikipedia.org/wiki/Spectrin__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSPVkUGws$ and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/ https://urldefense.com/v3/__https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSkI3UiIc$. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
ChimeraX-users mailing listChimeraX-users@cgl.ucsf.edu Manage subscription:https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSxjHBVGk$
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
Hi Roden,
Sure, no problem. But it will probably be a few weeks before I can get to it - have to get the release out first!
-- Tristan ________________________________ From: Roden Deng Luo deng.luo@kaust.edu.sa Sent: 26 October 2021 12:32 To: Tristan Croll tic20@cam.ac.uk Cc: goddard goddard@sonic.net; chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu Subject: Re: [chimerax-users] Refine predicted structure by pulling
Hi Tristan, Many thanks! I will get myself prepared for ISOLDE. If it is indeed possible to add in crosslinking and possible for multiple proteins, would it be fine for you to write a short description of how to incorporate the crosslinking data at the Python layer (not urgent at all, just at your convenience)? (It has been a while since I deviated from Python and dived into c++ and CUDA for graphics and parallel computing. But it should be fine with a quick review on Python syntax and then a few examples of achieving crosslinking between two demo structures).
Best, Roden
On Tue, Oct 26, 2021 at 2:10 PM Tristan Croll <tic20@cam.ac.ukmailto:tic20@cam.ac.uk> wrote: Hi Roden,
Glad you're happy!
1. ISOLDE is currently only capable of using a single GPU, so your big cluster isn't going to help much here. This was done on a laptop with a RTX 2080 GPU, and took about 20-30 minutes. 2. Try "usage isolde restrain", click one of the links that appears in the log, and you'll get the most up-to-date documentation for the command. I know these particular commands are a little on the complex side and I am trying to work out ways to make them more user-friendly, but persevere and you'll get the hang of it. Don't be afraid to experiment (just save your work first!). 3. In theory incorporation of chemical crosslinking data is possible, but right now I'm afraid it's only available through the Python layer (i.e. you'll need to do a little scripting). One of many things I'm hoping to flesh out in future. 4. There's no functionality at present to constrain within a mesh with hard boundaries, since everything in ISOLDE is gradient driven. You'd need to figure out a way to convert your mesh into some form of simulated map (i.e. with gradually increasing "density" values from the surface to the centre).
Best regards, Tristan
________________________________ From: Roden Deng Luo <deng.luo@kaust.edu.samailto:deng.luo@kaust.edu.sa> Sent: 26 October 2021 12:00 To: Tristan Croll <tic20@cam.ac.ukmailto:tic20@cam.ac.uk> Cc: goddard <goddard@sonic.netmailto:goddard@sonic.net>; chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu <chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu> Subject: Re: [chimerax-users] Refine predicted structure by pulling
Dear Tristan,
Thanks! I think this is exactly what I have been looking forward to for quite some time. A few follow-up questions.
1. What is your hardware and how long does it take to reach the state you showed? I got one RTX 3090 on my local workstation (dual boot for Win10 and Ubuntu). Additional four A6000 on a remote ubuntu server and V100s on a cluster. Do you think I would need to bother with setting up the workflow and X11 forwarding on the remote machines or one RTX 3090 is fine for daily tasks? 2. I feel intimidated by the jargons mentioned above. I found this documentation (https://isolde.cimr.cam.ac.uk/documentation/https://urldefense.com/v3/__https://isolde.cimr.cam.ac.uk/documentation/__;!!Nmw4Hv0!lzCpbVPCotIxWeqmSBW4kHWs5yl7AFeWsveYGcv3E0N6fpto92B_Byoio2SdfQcFrdM$). Apart from that, do you have other suggestions for reading materials? 3. Would it be possible to have multiple proteins in the scene and incorporate chemical crosslink data? It gives us the information of amino acid X should be within e.g. 30 Å to amino acid Y. 4. Would it be possible to constrain the structure inside a general mesh (e.g. obj file) rather than an EM density map?
Thanks, Roden
On Mon, Oct 25, 2021 at 11:37 PM Tristan Croll <tic20@cam.ac.ukmailto:tic20@cam.ac.uk> wrote: Result of some quite aggressive stretching (alternating pulling on residues 2000 onwards and 1-700, extending the tugged selection at both ends as things straightened out). This was with my dev ISOLDE build allowing me to use the "adjustForConfidence true" argument and an 8 Angstrom distance cutoff, which definitely helped keep things under control (also ended up restraining torsions with "isolde restrain torsions #1". Quick-and-dirty video overview of the final model at https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=s...https://urldefense.com/v3/__https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=sharing__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSS_JfJXg$, showing that almost all the distance restraints remain well satisfied, except for a few at the hinges.
Hope you find it useful!
Tristan
________________________________ From: ChimeraX-users <chimerax-users-bounces@cgl.ucsf.edumailto:chimerax-users-bounces@cgl.ucsf.edu> on behalf of Tristan Croll via ChimeraX-users <chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu> Sent: 25 October 2021 21:14 To: goddard <goddard@sonic.netmailto:goddard@sonic.net>; chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu <chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu>; Roden Deng Luo <deng.luo@kaust.edu.samailto:deng.luo@kaust.edu.sa> Subject: Re: [chimerax-users] Refine predicted structure by pulling
Hi Tom,
Yes, "kappa" is the stiffness. Default in the current release is 5, (as specified in the documentation via "usage isolde restrain") which I agree is very soft for this particular purpose - after playing around with it a bit I find "isolde restrain distances #1 distance 5 kappa 100 fallOff -3" works quite well. The extra "adjustForConfidence true" argument coming in the new ISOLDE release (not yet available - here in a few weeks) will help further - because it uses AlphaFold's own confidence on residue-residue distances, it can extend the distance restraints over a longer range (e.g. the default 8 Angstroms) while ignoring spurious chance contacts between residues. Will play around a little more and share the result.
-- Tristan ________________________________ From: Tom Goddard <goddard@sonic.netmailto:goddard@sonic.net> Sent: 25 October 2021 20:01 To: chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu <chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu>; Roden Deng Luo <deng.luo@kaust.edu.samailto:deng.luo@kaust.edu.sa>; Tristan Croll <tic20@cam.ac.ukmailto:tic20@cam.ac.uk> Subject: Re: [chimerax-users] Refine predicted structure by pulling
Hi Roden, Tristan,
I tried Tristan's suggestion for straightening the long AlphaFold helix model. It didn't work well for me because the restraints were too soft, so if I pulled on the first 160 residues it hardly moved the attached 2000 and instead just distorted the linker. Tristan is there a way to make the restraints much stiffer? I see the "isolde restrain distance" command has a "kappa" option and "falloff" option but I don't know if kappa is a stiffness or what magnitude it usually uses.
I think a better approach is to just rigidly rotate and translate large straight sections by hand using the "Move atoms" mouse mode in the ChimeraX 1.3 toolbar tab called Right Mouse. I tried this and have attached an image of about 10 minutes fiddling by hand. I hover the mouse at a hinge residue to get the residue number and then select everything on one side of that hinge "select :1182-2419" then use the right mouse button in the "move atoms" mode, holding the shift key down to do rotations, and no shift key to do translations. I zoom in on the hinge to try to adjust so the bond is not stretched and to avoid clashes. It would probably take 1 hour to do a good job. With my 10 minute test I then saved an mmCIF file and tried in ChimeraX 1.2.5 to minimize it with ISOLDE, but it said there were severe clashes.
Tom
On 10/24/2021 1:49 AM, Tristan Croll via ChimeraX-users wrote: Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin (https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisoldehttps://urldefense.com/v3/__https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS07ItNcc$). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
* first, check carefully for entanglement between distant domains (this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem. * Start ISOLDE. Then, assuming your protein is model #1, do the following commands: * isolde restrain distances #1 distanceCutoff 5 * isolde restrain torsions #1 angleRange 60 * isolde sim start #1 * Then, you'll want to use the "tug selection" mouse mode (third from bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle. * If you have unwanted distance restraints between domains that should be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel". * You might find it easier for a job like this to reduce the view to a C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
________________________________ From: ChimeraX-users chimerax-users-bounces@cgl.ucsf.edumailto:chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu Sent: 23 October 2021 16:30 To: chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edumailto:chimerax-users@cgl.ucsf.edu Subject: [chimerax-users] Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549https://urldefense.com/v3/__https://alphafold.ebi.ac.uk/entry/P02549__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS138lyq8$. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrinhttps://urldefense.com/v3/__https://en.wikipedia.org/wiki/Spectrin__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSPVkUGws$ and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/https://urldefense.com/v3/__https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSkI3UiIc$. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
________________________________ This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edumailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-usershttps://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSxjHBVGk$
________________________________ This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
________________________________ This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
Sure! Very appreciated. Good luck with the release.
On Tue, Oct 26, 2021 at 2:35 PM Tristan Croll tic20@cam.ac.uk wrote:
Hi Roden,
Sure, no problem. But it will probably be a few weeks before I can get to it - have to get the release out first!
-- Tristan
*From:* Roden Deng Luo deng.luo@kaust.edu.sa *Sent:* 26 October 2021 12:32 *To:* Tristan Croll tic20@cam.ac.uk *Cc:* goddard goddard@sonic.net; chimerax-users@cgl.ucsf.edu < chimerax-users@cgl.ucsf.edu> *Subject:* Re: [chimerax-users] Refine predicted structure by pulling
Hi Tristan, Many thanks! I will get myself prepared for ISOLDE. If it is indeed possible to add in crosslinking and possible for multiple proteins, would it be fine for you to write a short description of how to incorporate the crosslinking data at the Python layer (not urgent at all, just at your convenience)? (It has been a while since I deviated from Python and dived into c++ and CUDA for graphics and parallel computing. But it should be fine with a quick review on Python syntax and then a few examples of achieving crosslinking between two demo structures).
Best, Roden
On Tue, Oct 26, 2021 at 2:10 PM Tristan Croll tic20@cam.ac.uk wrote:
Hi Roden,
Glad you're happy!
- ISOLDE is currently only capable of using a single GPU, so your big
cluster isn't going to help much here. This was done on a laptop with a RTX 2080 GPU, and took about 20-30 minutes. 2. Try "usage isolde restrain", click one of the links that appears in the log, and you'll get the most up-to-date documentation for the command. I know these particular commands are a little on the complex side and I am trying to work out ways to make them more user-friendly, but persevere and you'll get the hang of it. Don't be afraid to experiment (just save your work first!). 3. In theory incorporation of chemical crosslinking data is possible, but right now I'm afraid it's only available through the Python layer (i.e. you'll need to do a little scripting). One of many things I'm hoping to flesh out in future. 4. There's no functionality at present to constrain within a mesh with hard boundaries, since everything in ISOLDE is gradient driven. You'd need to figure out a way to convert your mesh into some form of simulated map (i.e. with gradually increasing "density" values from the surface to the centre).
Best regards, Tristan
*From:* Roden Deng Luo deng.luo@kaust.edu.sa *Sent:* 26 October 2021 12:00 *To:* Tristan Croll tic20@cam.ac.uk *Cc:* goddard goddard@sonic.net; chimerax-users@cgl.ucsf.edu < chimerax-users@cgl.ucsf.edu> *Subject:* Re: [chimerax-users] Refine predicted structure by pulling
Dear Tristan,
Thanks! I think this is exactly what I have been looking forward to for quite some time. A few follow-up questions.
- What is your hardware and how long does it take to reach the state you
showed? I got one RTX 3090 on my local workstation (dual boot for Win10 and Ubuntu). Additional four A6000 on a remote ubuntu server and V100s on a cluster. Do you think I would need to bother with setting up the workflow and X11 forwarding on the remote machines or one RTX 3090 is fine for daily tasks? 2. I feel intimidated by the jargons mentioned above. I found this documentation (https://isolde.cimr.cam.ac.uk/documentation/ https://urldefense.com/v3/__https://isolde.cimr.cam.ac.uk/documentation/__;!!Nmw4Hv0!lzCpbVPCotIxWeqmSBW4kHWs5yl7AFeWsveYGcv3E0N6fpto92B_Byoio2SdfQcFrdM$). Apart from that, do you have other suggestions for reading materials? 3. Would it be possible to have multiple proteins in the scene and incorporate chemical crosslink data? It gives us the information of amino acid X should be within e.g. 30 Å to amino acid Y. 4. Would it be possible to constrain the structure inside a general mesh (e.g. obj file) rather than an EM density map?
Thanks, Roden
On Mon, Oct 25, 2021 at 11:37 PM Tristan Croll tic20@cam.ac.uk wrote:
Result of some quite aggressive stretching (alternating pulling on residues 2000 onwards and 1-700, extending the tugged selection at both ends as things straightened out). This was with my dev ISOLDE build allowing me to use the "adjustForConfidence true" argument and an 8 Angstrom distance cutoff, which definitely helped keep things under control (also ended up restraining torsions with "isolde restrain torsions #1". Quick-and-dirty video overview of the final model at https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=s... https://urldefense.com/v3/__https://drive.google.com/file/d/15zCt7WF7W6rBLoujMhFwgdpDYaPsYMB-/view?usp=sharing__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSS_JfJXg$, showing that almost all the distance restraints remain well satisfied, except for a few at the hinges.
Hope you find it useful!
Tristan
*From:* ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu on behalf of Tristan Croll via ChimeraX-users chimerax-users@cgl.ucsf.edu *Sent:* 25 October 2021 21:14 *To:* goddard goddard@sonic.net; chimerax-users@cgl.ucsf.edu < chimerax-users@cgl.ucsf.edu>; Roden Deng Luo deng.luo@kaust.edu.sa *Subject:* Re: [chimerax-users] Refine predicted structure by pulling
Hi Tom,
Yes, "kappa" is the stiffness. Default in the current release is 5, (as specified in the documentation via "usage isolde restrain") which I agree is very soft for this particular purpose - after playing around with it a bit I find "isolde restrain distances #1 distance 5 kappa 100 fallOff -3" works quite well. The extra "adjustForConfidence true" argument coming in the new ISOLDE release (not yet available - here in a few weeks) will help further - because it uses AlphaFold's own confidence on residue-residue distances, it can extend the distance restraints over a longer range (e.g. the default 8 Angstroms) while ignoring spurious chance contacts between residues. Will play around a little more and share the result.
-- Tristan
*From:* Tom Goddard goddard@sonic.net *Sent:* 25 October 2021 20:01 *To:* chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu; Roden Deng Luo deng.luo@kaust.edu.sa; Tristan Croll tic20@cam.ac.uk *Subject:* Re: [chimerax-users] Refine predicted structure by pulling
Hi Roden, Tristan,
I tried Tristan's suggestion for straightening the long AlphaFold helix model. It didn't work well for me because the restraints were too soft, so if I pulled on the first 160 residues it hardly moved the attached 2000 and instead just distorted the linker. Tristan is there a way to make the restraints much stiffer? I see the "isolde restrain distance" command has a "kappa" option and "falloff" option but I don't know if kappa is a stiffness or what magnitude it usually uses.
I think a better approach is to just rigidly rotate and translate large straight sections by hand using the "Move atoms" mouse mode in the ChimeraX 1.3 toolbar tab called Right Mouse. I tried this and have attached an image of about 10 minutes fiddling by hand. I hover the mouse at a hinge residue to get the residue number and then select everything on one side of that hinge "select :1182-2419" then use the right mouse button in the "move atoms" mode, holding the shift key down to do rotations, and no shift key to do translations. I zoom in on the hinge to try to adjust so the bond is not stretched and to avoid clashes. It would probably take 1 hour to do a good job. With my 10 minute test I then saved an mmCIF file and tried in ChimeraX 1.2.5 to minimize it with ISOLDE, but it said there were severe clashes.
TomOn 10/24/2021 1:49 AM, Tristan Croll via ChimeraX-users wrote:
Hi Roden,
This should be reasonably straightforward with the ISOLDE plugin ( https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde https://urldefense.com/v3/__https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxisolde__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS07ItNcc$). You'll also want to be working on a machine with a good GPU to make it feasible for a model this size. It'll be easier still with the new release coming in a few weeks, which has some added tools specifically for working with AlphaFold models. But with care you could do it now:
- first, check carefully for entanglement between distant domains
(this is an issue for AlphaFold with these very long, flexible proteins - if distant parts of the protein have no coevolutionary signal they don't really "know" about each other and can end up trying to occupy the same space). Looks like your particular example is free of that problem.
- Start ISOLDE. Then, assuming your protein is model #1, do the
following commands: - isolde restrain distances #1 distanceCutoff 5 - isolde restrain torsions #1 angleRange 60 - isolde sim start #1
- Then, you'll want to use the "tug selection" mouse mode (third from
bottom right of the ISOLDE panel). Make a selection for a suitable region to tug (e.g. "sel #1-100@CA" to tug the N-terminus), and tugging forces applied by right-click-and-drag will be spread over that selection. Keep a close eye on the distance restraints - if they start to become too strained (lots of purple appearing), stop tugging and let the model settle.
- If you have unwanted distance restraints between domains that should
be distant from each other, you can selectively release restraints on selected atoms with "isolde release distances sel".
- You might find it easier for a job like this to reduce the view to a
C-alpha trace - just "hide ~@CA" - if you do this after you've started a simulation, it'll automatically revert to the all-atom view once you stop.
Happy modelling!
-- Tristan
*From:* ChimeraX-users chimerax-users-bounces@cgl.ucsf.edu chimerax-users-bounces@cgl.ucsf.edu on behalf of Roden Deng Luo via ChimeraX-users chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu *Sent:* 23 October 2021 16:30 *To:* chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu chimerax-users@cgl.ucsf.edu *Subject:* [chimerax-users] Refine predicted structure by pulling
Dear ChimeraX Users,
There is this AlphaFold prediction: https://alphafold.ebi.ac.uk/entry/P02549 https://urldefense.com/v3/__https://alphafold.ebi.ac.uk/entry/P02549__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkS138lyq8$. And we know this protein should be more or less in a rod-like shape rather than a globular shape: https://en.wikipedia.org/wiki/Spectrin https://urldefense.com/v3/__https://en.wikipedia.org/wiki/Spectrin__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSPVkUGws$ and related literature such as https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/ https://urldefense.com/v3/__https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3218374/__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSkI3UiIc$. I wonder is there a way that we can pull the two ends of the predicted structure while maintaining the secondary structures, mostly alpha-helices in this case, inside ChimeraX?
I am quite new to ChimeraX. I understand this is more fitted to an MD simulation. But I got to know there are options to fit a PDB to a density map. So I wondered if this can be achieved in ChimeraX.
Best, Roden
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
ChimeraX-users mailing listChimeraX-users@cgl.ucsf.edu Manage subscription:https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!lCB3QYi96tlHwoqUxwqr0KVeJnh1_mPFwsjQ89mTgsUb6GWyfULSELF3pDkSxjHBVGk$
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
participants (3)
-
Roden Deng Luo -
Tom Goddard -
Tristan Croll