- Posts: 13
Parameterizing an amphiphilic polymer
- ywang52
-
Topic Author
- Offline
- Fresh Boarder
Less
More
7 years 8 months ago #5774
by ywang52
Parameterizing an amphiphilic polymer was created by ywang52
Hi everybody,
I am trying to make a MARTINI model for an amphiphilic homopolymer (each monomer has a hydrophobic chain and a hydrophilic chain on it). I am now doing the parameterization according to the procedure given in the tutorial. I am really confused now by some problem.
The way I did it was that I ran the atomistic simulation first for the monomer in water at normal T and P and get the bond length and angle distribution from the mapped trajectory. Then I created a CG topology accordingly for the monomer and ran it in water again in the same condition. In this CG topology I did not care about dihedral yet, only bonds, constraints and angles were described. After about 20 times trial and error process, most of the distributions reached good agreement, but not all of them. The monomer contains two aromatic rings in it. All those distributions which did not reach good agreement are related to the ring. I modeled the ring by three SC4 beads. I mentioned the ring structure not because I think they are the problem. I am really not sure about it. I have couple of questions,
1) In the CG monomer simulation, I used standard MARTINI water, could it be the reason of the unmatched distributions? I am now running another simulation with polarizable water model, I don't know if it will help.
2) The model will eventually be used in the polymer system in different solvent, for example in benzene. In this case, do I have to parameterize the molecule again from the monomer-benzene atomistic system? What if with different T or P?
3) For now, my understand is that we usually don't care much about dihedrals in the CG model unless it is required for some specific case. Most of the dihedral distributions I got from the mapped atomistic simulation seem not have much rule, am I right?
Thanks very much, any help will be appreciated.
Best
Yang
I am trying to make a MARTINI model for an amphiphilic homopolymer (each monomer has a hydrophobic chain and a hydrophilic chain on it). I am now doing the parameterization according to the procedure given in the tutorial. I am really confused now by some problem.
The way I did it was that I ran the atomistic simulation first for the monomer in water at normal T and P and get the bond length and angle distribution from the mapped trajectory. Then I created a CG topology accordingly for the monomer and ran it in water again in the same condition. In this CG topology I did not care about dihedral yet, only bonds, constraints and angles were described. After about 20 times trial and error process, most of the distributions reached good agreement, but not all of them. The monomer contains two aromatic rings in it. All those distributions which did not reach good agreement are related to the ring. I modeled the ring by three SC4 beads. I mentioned the ring structure not because I think they are the problem. I am really not sure about it. I have couple of questions,
1) In the CG monomer simulation, I used standard MARTINI water, could it be the reason of the unmatched distributions? I am now running another simulation with polarizable water model, I don't know if it will help.
2) The model will eventually be used in the polymer system in different solvent, for example in benzene. In this case, do I have to parameterize the molecule again from the monomer-benzene atomistic system? What if with different T or P?
3) For now, my understand is that we usually don't care much about dihedrals in the CG model unless it is required for some specific case. Most of the dihedral distributions I got from the mapped atomistic simulation seem not have much rule, am I right?
Thanks very much, any help will be appreciated.
Best
Yang
Please Log in or Create an account to join the conversation.
- xavier
-
- Offline
- Admin
Less
More
- Posts: 416
7 years 8 months ago #5775
by xavier
Overall you seem to be on the right track, Although it is important to reproduce distributions of bonds/angle/dihedrals it might not be possible to do it for all of them. The most important is to reproduce the conformational flexility at best possible considering the CG limitations.
Good luck.
Replied by xavier on topic Parameterizing an amphiphilic polymer
It is very unlikely but the use of the polarisable water might help getting better electrostatic screening if it is important.ywang52 wrote: Hi everybody,
I am trying to make a MARTINI model for an amphiphilic homopolymer (each monomer has a hydrophobic chain and a hydrophilic chain on it). I am now doing the parameterization according to the procedure given in the tutorial. I am really confused now by some problem.
The way I did it was that I ran the atomistic simulation first for the monomer in water at normal T and P and get the bond length and angle distribution from the mapped trajectory. Then I created a CG topology accordingly for the monomer and ran it in water again in the same condition. In this CG topology I did not care about dihedral yet, only bonds, constraints and angles were described. After about 20 times trial and error process, most of the distributions reached good agreement, but not all of them. The monomer contains two aromatic rings in it. All those distributions which did not reach good agreement are related to the ring. I modeled the ring by three SC4 beads. I mentioned the ring structure not because I think they are the problem. I am really not sure about it. I have couple of questions,
1) In the CG monomer simulation, I used standard MARTINI water, could it be the reason of the unmatched distributions? I am now running another simulation with polarizable water model, I don't know if it will help.
It is important that you do indeed check that the model of the monomer (and polymer) explores similar conformational space in both solvent. Similarly for T and P if the changes in P and T are significant.2) The model will eventually be used in the polymer system in different solvent, for example in benzene. In this case, do I have to parameterize the molecule again from the monomer-benzene atomistic system? What if with different T or P?
Dihedrals might indeed be useful to describe a certain conformations and we do care that as bonds and angle they have to be reasonably reproduced.3) For now, my understand is that we usually don't care much about dihedrals in the CG model unless it is required for some specific case. Most of the dihedral distributions I got from the mapped atomistic simulation seem not have much rule, am I right?
Overall you seem to be on the right track, Although it is important to reproduce distributions of bonds/angle/dihedrals it might not be possible to do it for all of them. The most important is to reproduce the conformational flexility at best possible considering the CG limitations.
Good luck.
Please Log in or Create an account to join the conversation.
- ywang52
-
Topic Author
- Offline
- Fresh Boarder
Less
More
- Posts: 13
7 years 8 months ago #5780
by ywang52
Thank you Xavier, your answer is really helpful.
Replied by ywang52 on topic Parameterizing an amphiphilic polymer
xavier wrote:
It is very unlikely but the use of the polarisable water might help getting better electrostatic screening if it is important.ywang52 wrote: Hi everybody,
I am trying to make a MARTINI model for an amphiphilic homopolymer (each monomer has a hydrophobic chain and a hydrophilic chain on it). I am now doing the parameterization according to the procedure given in the tutorial. I am really confused now by some problem.
The way I did it was that I ran the atomistic simulation first for the monomer in water at normal T and P and get the bond length and angle distribution from the mapped trajectory. Then I created a CG topology accordingly for the monomer and ran it in water again in the same condition. In this CG topology I did not care about dihedral yet, only bonds, constraints and angles were described. After about 20 times trial and error process, most of the distributions reached good agreement, but not all of them. The monomer contains two aromatic rings in it. All those distributions which did not reach good agreement are related to the ring. I modeled the ring by three SC4 beads. I mentioned the ring structure not because I think they are the problem. I am really not sure about it. I have couple of questions,
1) In the CG monomer simulation, I used standard MARTINI water, could it be the reason of the unmatched distributions? I am now running another simulation with polarizable water model, I don't know if it will help.It is important that you do indeed check that the model of the monomer (and polymer) explores similar conformational space in both solvent. Similarly for T and P if the changes in P and T are significant.2) The model will eventually be used in the polymer system in different solvent, for example in benzene. In this case, do I have to parameterize the molecule again from the monomer-benzene atomistic system? What if with different T or P?
Dihedrals might indeed be useful to describe a certain conformations and we do care that as bonds and angle they have to be reasonably reproduced.3) For now, my understand is that we usually don't care much about dihedrals in the CG model unless it is required for some specific case. Most of the dihedral distributions I got from the mapped atomistic simulation seem not have much rule, am I right?
Overall you seem to be on the right track, Although it is important to reproduce distributions of bonds/angle/dihedrals it might not be possible to do it for all of them. The most important is to reproduce the conformational flexility at best possible considering the CG limitations.
Good luck.
Thank you Xavier, your answer is really helpful.
Please Log in or Create an account to join the conversation.
- ywang52
-
Topic Author
- Offline
- Fresh Boarder
Less
More
- Posts: 13
7 years 8 months ago #5786
by ywang52
Replied by ywang52 on topic Parameterizing an amphiphilic polymer
Hi Xavier
I just got some new questionss regarding to the parameterizing process.
I am confused about what should I do if the distribution obtained from CG simulation shifts. For example, I have a value of bond length as 0.4 nm obtained from the mapped all atom trajectory, then I use this value in the topology of my CG model, then I get the distribution from the CG simulation. However, the distribution shifts, and give me a value of 0.5 nm. In this case, can I just simply change the value from 0.4 nm to, say, 0.3 nm, which is not the value given by all atom simulation but makes the CG distribution matches the all atom distribution?
If not, should I change the non-bonded parameters or considering another type of bead to represent the model?
Thanks in advance,
Yang
I just got some new questionss regarding to the parameterizing process.
I am confused about what should I do if the distribution obtained from CG simulation shifts. For example, I have a value of bond length as 0.4 nm obtained from the mapped all atom trajectory, then I use this value in the topology of my CG model, then I get the distribution from the CG simulation. However, the distribution shifts, and give me a value of 0.5 nm. In this case, can I just simply change the value from 0.4 nm to, say, 0.3 nm, which is not the value given by all atom simulation but makes the CG distribution matches the all atom distribution?
If not, should I change the non-bonded parameters or considering another type of bead to represent the model?
Thanks in advance,
Yang
Please Log in or Create an account to join the conversation.
- flaviyan
-
- Offline
- Senior Boarder
Less
More
- Posts: 54
7 years 8 months ago #5788
by flaviyan
Replied by flaviyan on topic Parameterizing an amphiphilic polymer
the force constant determines the SD of the distribution so you can try increasing the force constant.
Please Log in or Create an account to join the conversation.
- ywang52
-
Topic Author
- Offline
- Fresh Boarder
Less
More
- Posts: 13
7 years 8 months ago #5789
by ywang52
Hi, thank you very much for your answer, but doesn't the force constant mainly control the width and height of the distribution? My problem now is that some distributions shift.
Replied by ywang52 on topic Parameterizing an amphiphilic polymer
flaviyan wrote: the force constant determines the SD of the distribution so you can try increasing the force constant.
Hi, thank you very much for your answer, but doesn't the force constant mainly control the width and height of the distribution? My problem now is that some distributions shift.
Please Log in or Create an account to join the conversation.
- flaviyan
-
- Offline
- Senior Boarder
Less
More
- Posts: 54
7 years 8 months ago #5790
by flaviyan
Replied by flaviyan on topic Parameterizing an amphiphilic polymer
You are right. if you set the equilibrium bond length at 0.4 nm and set a force constant as lets say 100 KJ/mol/nm2 then you are giving more flexibility to the bond and your peaks probably would shift to the optimal LJ distance of the two beads in question, but if you increase the force constant to lets say 10000 KJ/mol/nm2 then of course your peak height and width would be narrower and as a consequence you will be more closer to the set bond length.
Please Log in or Create an account to join the conversation.
- ywang52
-
Topic Author
- Offline
- Fresh Boarder
Less
More
- Posts: 13
7 years 8 months ago #5791
by ywang52
Thank you for your explanation. I will try and see what will happen.
Replied by ywang52 on topic Parameterizing an amphiphilic polymer
flaviyan wrote: You are right. if you set the equilibrium bond length at 0.4 nm and set a force constant as lets say 100 KJ/mol/nm2 then you are giving more flexibility to the bond and your peaks probably would shift to the optimal LJ distance of the two beads in question, but if you increase the force constant to lets say 10000 KJ/mol/nm2 then of course your peak height and width would be narrower and as a consequence you will be more closer to the set bond length.
Thank you for your explanation. I will try and see what will happen.
Please Log in or Create an account to join the conversation.
- peterkroon
-
- Offline
- Gold Boarder
Less
More
- Posts: 210
7 years 8 months ago #5793
by peterkroon
Replied by peterkroon on topic Parameterizing an amphiphilic polymer
The goal is that you reproduce the atomistic distributions, so feel free to lower the average bond lengths. However, be aware that these discrepancies between .top file and found distribution can result in large forces and unstable simulations.
Further, be somewhat careful with too short bond lengths, since bead-bead overlap can cause problems with the non-bonded parameters.
Further, be somewhat careful with too short bond lengths, since bead-bead overlap can cause problems with the non-bonded parameters.
Please Log in or Create an account to join the conversation.
- xavier
-
- Offline
- Admin
Less
More
- Posts: 416
7 years 7 months ago #5834
by xavier
Replied by xavier on topic Parameterizing an amphiphilic polymer
Sorry for the delay in answering.
If your CG distribution does not match the AT one, the first thing you want to do is increase the force constant of the bond. You can go up to 30000-50000 with a time step of 20fs. But be aware that these are relative strong force constant.
Reducing the ref distance is not what you want to do. It would be similar to increase the force constant anyways :)).
Note that if the CG distribution is not reproduced, it might be due to other interactions of the two beads with the rest of the molecule. You may want to look at these and potentially exclude some (mostly non-bonded) interactions that are interfering with reproducing a particular distance and would not be relevant for maintaining the module's structure.
Note: two bonded atoms will not feel the no-bonded (Elect/vdW) interactions as they are excluded from the list.
If your CG distribution does not match the AT one, the first thing you want to do is increase the force constant of the bond. You can go up to 30000-50000 with a time step of 20fs. But be aware that these are relative strong force constant.
Reducing the ref distance is not what you want to do. It would be similar to increase the force constant anyways :)).
Note that if the CG distribution is not reproduced, it might be due to other interactions of the two beads with the rest of the molecule. You may want to look at these and potentially exclude some (mostly non-bonded) interactions that are interfering with reproducing a particular distance and would not be relevant for maintaining the module's structure.
Note: two bonded atoms will not feel the no-bonded (Elect/vdW) interactions as they are excluded from the list.
Please Log in or Create an account to join the conversation.
- ywang52
-
Topic Author
- Offline
- Fresh Boarder
Less
More
- Posts: 13
7 years 7 months ago #5841
by ywang52
Thanks Xavier
I really appreciate your answer! However, I still have some confusions.
1. For very sharp distribution from AA simulation, do I have to reproduce it exactly?
For example, the bond length distributions of a ring structure is usually very sharp, the way I tried to do it was that I used a very very large force constant (100000 ~ 200000, and of course I could not use 20 fs time step anymore), which can help me to reproduce both the height and the width of the distribution. (constraint give me a way sharper distribution).
So, are you saying that we don't need to care about how "fat" the distributions are and we just need to make sure the peak of the distributions is consistent with the reference value we assigned?
Also, if we use constraint for some bond, can we still use Lincs?
2. How can I identify the possible nonbonded interaction interfering with reproducing distributions? Is there any procedure that I can follow?
3. Do you recommend changing the nonbonded parameters of beads, I did not do it, because that is basically creating new beads in my mind.
Thank you very much Xavier!
Replied by ywang52 on topic Parameterizing an amphiphilic polymer
xavier wrote: Sorry for the delay in answering.
If your CG distribution does not match the AT one, the first thing you want to do is increase the force constant of the bond. You can go up to 30000-50000 with a time step of 20fs. But be aware that these are relative strong force constant.
Reducing the ref distance is not what you want to do. It would be similar to increase the force constant anyways :)).
Note that if the CG distribution is not reproduced, it might be due to other interactions of the two beads with the rest of the molecule. You may want to look at these and potentially exclude some (mostly non-bonded) interactions that are interfering with reproducing a particular distance and would not be relevant for maintaining the module's structure.
Note: two bonded atoms will not feel the no-bonded (Elect/vdW) interactions as they are excluded from the list.
Thanks Xavier
I really appreciate your answer! However, I still have some confusions.
1. For very sharp distribution from AA simulation, do I have to reproduce it exactly?
For example, the bond length distributions of a ring structure is usually very sharp, the way I tried to do it was that I used a very very large force constant (100000 ~ 200000, and of course I could not use 20 fs time step anymore), which can help me to reproduce both the height and the width of the distribution. (constraint give me a way sharper distribution).
So, are you saying that we don't need to care about how "fat" the distributions are and we just need to make sure the peak of the distributions is consistent with the reference value we assigned?
Also, if we use constraint for some bond, can we still use Lincs?
2. How can I identify the possible nonbonded interaction interfering with reproducing distributions? Is there any procedure that I can follow?
3. Do you recommend changing the nonbonded parameters of beads, I did not do it, because that is basically creating new beads in my mind.
Thank you very much Xavier!
Please Log in or Create an account to join the conversation.
- xavier
-
- Offline
- Admin
Less
More
- Posts: 416
7 years 7 months ago #5842
by xavier
1) For such narrow distributions we do recommend using constrains. This is especially true for ring structures.
2) Unfortunately there are no "automated" procedure. You have to visualise your system and look at neighbouring atoms. It should come obvious quickly, some trail and errors would speed up the process.
3) Changing the non-bonded to make the bonded better match atomistic distributions would not be a good idea! The main option you have is to exclude some undesirable non-bonded interactions.
I am happy if this helps.
XA+
Replied by xavier on topic Parameterizing an amphiphilic polymer
ywang52 wrote:
xavier wrote: Sorry for the delay in answering.
If your CG distribution does not match the AT one, the first thing you want to do is increase the force constant of the bond. You can go up to 30000-50000 with a time step of 20fs. But be aware that these are relative strong force constant.
Reducing the ref distance is not what you want to do. It would be similar to increase the force constant anyways :)).
Note that if the CG distribution is not reproduced, it might be due to other interactions of the two beads with the rest of the molecule. You may want to look at these and potentially exclude some (mostly non-bonded) interactions that are interfering with reproducing a particular distance and would not be relevant for maintaining the module's structure.
Note: two bonded atoms will not feel the no-bonded (Elect/vdW) interactions as they are excluded from the list.
Thanks Xavier
I really appreciate your answer! However, I still have some confusions.
1. For very sharp distribution from AA simulation, do I have to reproduce it exactly?
For example, the bond length distributions of a ring structure is usually very sharp, the way I tried to do it was that I used a very very large force constant (100000 ~ 200000, and of course I could not use 20 fs time step anymore), which can help me to reproduce both the height and the width of the distribution. (constraint give me a way sharper distribution).
So, are you saying that we don't need to care about how "fat" the distributions are and we just need to make sure the peak of the distributions is consistent with the reference value we assigned?
Also, if we use constraint for some bond, can we still use Lincs?
2. How can I identify the possible nonbonded interaction interfering with reproducing distributions? Is there any procedure that I can follow?
3. Do you recommend changing the nonbonded parameters of beads, I did not do it, because that is basically creating new beads in my mind.
Thank you very much Xavier!
1) For such narrow distributions we do recommend using constrains. This is especially true for ring structures.
2) Unfortunately there are no "automated" procedure. You have to visualise your system and look at neighbouring atoms. It should come obvious quickly, some trail and errors would speed up the process.
3) Changing the non-bonded to make the bonded better match atomistic distributions would not be a good idea! The main option you have is to exclude some undesirable non-bonded interactions.
I am happy if this helps.
XA+
Please Log in or Create an account to join the conversation.
Time to create page: 0.118 seconds