- Posts: 24
LINCS warning in CG energy minimisation
- acnash
-
Topic Author
- Offline
- Junior Boarder
Less
More
7 years 3 months ago #6862
by acnash
Replied by acnash on topic LINCS warning in CG energy minimisation
Hi Thanks, for the reply. Hope you had a good new year. I spent most of my time looking into this issue, and as I began to understand a lot more about my system and how Martini was treating it I decided to open a new thread. I believe the issue I am having is being caused by the time-step's integration of the placement of ringed sidechain atoms (HIE, TYR etc...). Not, the backbone to sidechain bead rigid bond, but the actual constraint between ring side chain beads. Unfortunately, I am not sure how to approach this problem from here.
Please Log in or Create an account to join the conversation.
- jbarnoud
-
- Offline
- Admin
Less
More
- Posts: 23
7 years 3 months ago #6863
by jbarnoud
Replied by jbarnoud on topic LINCS warning in CG energy minimisation
Hi,
I may have missed it, but did you try to run the EM with non-polarizable water? I expect the lower density of constraints to make it easier to the minimization. If it works, there is a script in the download section of the site that transforms normal water to polarizable one. You will need one more EM after you transformed the water, but there will be less to minimize.
Also, minimization does not deal well with constraints. One trick is to use stiff bonds instead of constraints, just for the EM. If I well remember, martinize produce an ITP with a "FLEXIBLE" part. Use "define = -DFLEXIBLE" in your mdp file to use it.
Finally, the CG representation of your protein is bulkier than the atomistic one. Maybe you have steric clashes between chains of your protein. You could use VMD to see if you have such clashes. EM should deal with them, but if the chains are really to close, you can get so high energies that EM is not enough.
Hopefully it helps.
I may have missed it, but did you try to run the EM with non-polarizable water? I expect the lower density of constraints to make it easier to the minimization. If it works, there is a script in the download section of the site that transforms normal water to polarizable one. You will need one more EM after you transformed the water, but there will be less to minimize.
Also, minimization does not deal well with constraints. One trick is to use stiff bonds instead of constraints, just for the EM. If I well remember, martinize produce an ITP with a "FLEXIBLE" part. Use "define = -DFLEXIBLE" in your mdp file to use it.
Finally, the CG representation of your protein is bulkier than the atomistic one. Maybe you have steric clashes between chains of your protein. You could use VMD to see if you have such clashes. EM should deal with them, but if the chains are really to close, you can get so high energies that EM is not enough.
Hopefully it helps.
Please Log in or Create an account to join the conversation.
- acnash
-
Topic Author
- Offline
- Junior Boarder
Less
More
- Posts: 24
7 years 3 months ago #6877
by acnash
Replied by acnash on topic LINCS warning in CG energy minimisation
Hi both,
After the very useful suggestions I went back to the basic single protein model (CG) in vacuum, rather than the pentra-fibril with water. Given that I took the structure from an atomistic structure and I was unable to energy minimise the CG derivation I figure that there are some awful overlaps. Understanding that rigid bonds help with minimisation (I didn't earlier on), I replaced all the constraints with a rigid bond. This now allows me to minimise the structure to a FMax of around 1200. I feel like this is getting somewhere.
I switched back to constraints and performed an NVT with a dt of 0.02. LINCS errors.
I switched to rigid bonds again, NAN temp, pressure, etc... Big crash.
I switched back to constraints and performed an NVT with a dt of 0.002, much like an atomistic simulation. No problem at all. I kept the velocities and switched back to 0.02. Still it didn't work.
When I investigate what is blowing up it is always the sidechains of ring amino acids e.g., the SC beads of PHE. If I set the dt to 0.005, which gets me a bit of a trajectory until LINCS errors are thrown, I can see that the constraints between the ring beads of the ringed amino acids are 'perfect' as to their description in the .itp file. There is no deviation. LINCS is doing its job. But then the very last frame, they blow up. Interestingly, the offending sidechain is not interacting directly with anything. It is facing out into the vacuum.
I decided to then take one of the offending side chains and run it on its own using the constraint, bond, bond angle and improper dihedral information, along with the energy minimised structure. A single amino acid in space. This works perfectly with 0.025 ps.
In summary.
I am able to perform an energy minimisation successfully over a system of fixed bonds.
The beads of sidechains adopt the correct eq distance after minimisation.
I am able to run a small time step (0.002) over a system of constraints. Any greater and it breaks.
Every LINCs error (numerous times, different starting velocities and temps), denote the beads assigned to sidechains of ring-amino acids.
Yet, an offending sidechain on its own is fine.
After the very useful suggestions I went back to the basic single protein model (CG) in vacuum, rather than the pentra-fibril with water. Given that I took the structure from an atomistic structure and I was unable to energy minimise the CG derivation I figure that there are some awful overlaps. Understanding that rigid bonds help with minimisation (I didn't earlier on), I replaced all the constraints with a rigid bond. This now allows me to minimise the structure to a FMax of around 1200. I feel like this is getting somewhere.
I switched back to constraints and performed an NVT with a dt of 0.02. LINCS errors.
I switched to rigid bonds again, NAN temp, pressure, etc... Big crash.
I switched back to constraints and performed an NVT with a dt of 0.002, much like an atomistic simulation. No problem at all. I kept the velocities and switched back to 0.02. Still it didn't work.
When I investigate what is blowing up it is always the sidechains of ring amino acids e.g., the SC beads of PHE. If I set the dt to 0.005, which gets me a bit of a trajectory until LINCS errors are thrown, I can see that the constraints between the ring beads of the ringed amino acids are 'perfect' as to their description in the .itp file. There is no deviation. LINCS is doing its job. But then the very last frame, they blow up. Interestingly, the offending sidechain is not interacting directly with anything. It is facing out into the vacuum.
I decided to then take one of the offending side chains and run it on its own using the constraint, bond, bond angle and improper dihedral information, along with the energy minimised structure. A single amino acid in space. This works perfectly with 0.025 ps.
In summary.
I am able to perform an energy minimisation successfully over a system of fixed bonds.
The beads of sidechains adopt the correct eq distance after minimisation.
I am able to run a small time step (0.002) over a system of constraints. Any greater and it breaks.
Every LINCs error (numerous times, different starting velocities and temps), denote the beads assigned to sidechains of ring-amino acids.
Yet, an offending sidechain on its own is fine.
Please Log in or Create an account to join the conversation.
- acnash
-
Topic Author
- Offline
- Junior Boarder
Less
More
- Posts: 24
7 years 3 months ago #6898
by acnash
Replied by acnash on topic LINCS warning in CG energy minimisation
I've taken a huge step back and thrown the atomistic representation out of the window. I've taken an 'idealised' collagen structure (not a crystal structure) and used the -collagen flag on the martinise python script, using Martini2.2.
Something I notice immediately, I'm able to perform an energy minimisation to less than <300, unlike my earlier structure which got nowhere near this force threshold.
However, what is really confusing is that I'm getting the same problem. LINCs errors due to constraint issues, and it is always with the constraints between sidechain beads of ring-like sidechains e.g., HIS.
Step 1930, time 48.25 (ps) LINCS WARNING
relative constraint deviation after LINCS:
rms nan, max 0.000002 (between atoms 1153 and 1154)
bonds that rotated more than 90 degrees:
atom 1 atom 2 angle previous, current, constraint length
I performed an energy minimisation using hard bonds then switched over to constraints for a short NVT at 0.025 ps timesteps.
Something I notice immediately, I'm able to perform an energy minimisation to less than <300, unlike my earlier structure which got nowhere near this force threshold.
However, what is really confusing is that I'm getting the same problem. LINCs errors due to constraint issues, and it is always with the constraints between sidechain beads of ring-like sidechains e.g., HIS.
Step 1930, time 48.25 (ps) LINCS WARNING
relative constraint deviation after LINCS:
rms nan, max 0.000002 (between atoms 1153 and 1154)
bonds that rotated more than 90 degrees:
atom 1 atom 2 angle previous, current, constraint length
I performed an energy minimisation using hard bonds then switched over to constraints for a short NVT at 0.025 ps timesteps.
Please Log in or Create an account to join the conversation.
- jbarnoud
-
- Offline
- Admin
Less
More
- Posts: 23
7 years 3 months ago #6899
by jbarnoud
Replied by jbarnoud on topic LINCS warning in CG energy minimisation
One way to understand the cause of such issue is to save every frame close to the crash. This way you can see what happens just before the crash.
Are you using an elastic network? For such a large structure as collagen, you probably should.
Are you using an elastic network? For such a large structure as collagen, you probably should.
Please Log in or Create an account to join the conversation.
- acnash
-
Topic Author
- Offline
- Junior Boarder
Less
More
- Posts: 24
7 years 3 months ago #6907
by acnash
Replied by acnash on topic LINCS warning in CG energy minimisation
Hi jbarnoud, as with an all atom simulation I've tried watching every frame up to the crash. I'll admit, it's quite tricky to understand what's going on in a bead-like representation, but essentially the backbone 'snaps'.
I've yet to try an elastic network. Thankfully, after two months of hammering away at this, going from an atomistic to a CG structure has become a well-versed art. I'll give it a go and let you know how it gets on.
I've yet to try an elastic network. Thankfully, after two months of hammering away at this, going from an atomistic to a CG structure has become a well-versed art. I'll give it a go and let you know how it gets on.
Please Log in or Create an account to join the conversation.
- gupta.rakesh2
-
- Offline
- Fresh Boarder
Less
More
- Posts: 5
6 years 11 months ago #7257
by gupta.rakesh2
Replied by gupta.rakesh2 on topic LINCS warning in CG energy minimisation
I am also facing similar problem in setting up the collagen simulation in water.
I have tried all kind of combinations as mentioned above in the mailing thread.
Can someone help me in it.
I have tried all kind of combinations as mentioned above in the mailing thread.
Can someone help me in it.
Please Log in or Create an account to join the conversation.
Time to create page: 0.133 seconds