The modeling of 1microsecond of short RNA in aqueous solution is possible if you will use the massively parallel computational code (Gromacs, NAMD, ...). The choice of the force field rely mostly on the events you expect to observe. No need to know the X-ray structure since the dynamics of the molecule will be very important during this 1us. However, you need to know exactly the sequence and supposed bonding between bases.

I have only made protein MD, but what is literature telling you? What you're trying to do? Folding? Conformational sampling? Some specific binding/reaction? Did anybody try to do what you want to do? What force fields/methods are they using?

You can probably use a server like this http://bioinformatics.oxfordjournals.org/content/24/17/1951.short to get a reasonable starting structure and then I'd use GROMACS with the AMBER force field. But if you've never done molecular dynamics, first thing I'd do is trying to reproduce some other MD study and go from there.

The modeling of 1microsecond of short RNA in aqueous solution is possible if you will use the massively parallel computational code (Gromacs, NAMD, ...). The choice of the force field rely mostly on the events you expect to observe. No need to know the X-ray structure since the dynamics of the molecule will be very important during this 1us. However, you need to know exactly the sequence and supposed bonding between bases.

As Massimo indicated, you should probably start with simulating short chain sequences of nucleic acids (choose a force field which has been used in literature extensively and reproduced the structure and thermodynamics, say AMBER) and then try to build your sequence of RNA. Hope the attachment helps.

http://www-ibmc.u-strasbg.fr/arn/Westhof/publ_West/bib2006/r2006_PAuffinger_jsponer.pdf

To get started on this project, you need to think about the science you want to understand - what do you want to find out about your molecule.

Until you do that, you will find it difficult to decide how to go about the calculation.

In the meantime, some comments about your system - for an all-atom, explicit solvent MD, 1 microsecond is a long time and will take a long time to simulate unless your molecule is very small and you can parrallelise the calculation to a large extent. MD is hugely dependent on the starting structure you use and at the all-atom level is very poor at crossing from one minimum to another so make sure you have some evidence for your starting structure.

As Massimo suggests, GROMACS/AMBER is a good system for simlations of nucleic acids and used to b provided free with some Linux distros (SuSE from what I remember)

Hi there! Are you sure it has to be all-atom? What is the problem under study? It could be easier and much much faster to do it in a coarse-grained way. With the force-fields -- be careful, you can't change the temperature, since force field parameters are tuned for a specific temperature, so, literally speaking, you would not have physical grounds to use them at any other temperature. But first: what is the phenomenon you are going to study? Using packages like AMBER or GROMACS could be an alternative to writing your own code, since there are coarse-grained versions as well (to the best of my knowledge). If using them, I would strongly suggest to read the manuals until you get a very good understanding, what does the code exactly do, how does it model the system, up to looking through the lines of the code (if available). And the most important -- think twice and get confident about the physical picture and model behind the code and only then run in. Wishing you a good luck! May bugs in the code bypass your study :-)

All this really depends on the question you are asking.

As Sandal said above. The nature of the help you need does depend on your question. What do you want to do?

Note, even with the x-ray/nmr data you are going to run into a tonne of problems. See homology requires structures too, and if you dont have structures, homology basically cant help you much. The RNA is somewhat similar to DNA so maybe if you can get your hands on structural data on it from some place you may have a starting point to maybe map your RNA sequence onto it.

A second step would be to ensure what sort of modifications happen to it. e.g. i read a study where they talk about modification to the RNA. You need to keep those modifications in mind. Once you have done all this you are now setup to go into the simulation stage. For simulation you require a specific forcefield, i use charmm27 all atom forcefield, but i have never tried with RNA so cant say much on that. However, before you simulate you really need to phrase the right questions. MD simulations cant be run unless you are very specific in those terms. A few things you might need in case you phrase any question is know how to go through various stages on an MD. For that you need to do an extensive literature review.

During that as Saldo suggested, go through the process of conducting MD.

On a seperate note i would like to add that someone always begins at the first step, before his work can lead to more doors. Dont be scared to try new things just because its too risky, you will never know unless you try it.

Best,

/A

If there is no crystal structure I guess you will use the rna-fold of zucker's lab to predict the folding?

I think you can use the secondary structures guessed by this soft to build distance/angle restraints (such as Watson-Crick, helicoidal parameters..). Then you may use simulated annealing simulation to predict possible structures which fit the restraints. With these starting structures in hand, you then will be able to relax them in explicit solvent. If several structures converge to an unique one, you may validate this final and reliable model. However, I am not sure that 1 microsecond is required.

Hi. I am not familiar with the modelling of RNA, but the molecules I model also do not have any structural experimental data that I could start from, so I know where you're coming from.

What I would suggest is to do a very good search of the literature to look for pieces of data that you could use to compare your model to, and in this way check its accuracy. The kind of data you look for will depend on the system you model, so you would need to have some idea of the kind of simulations you need to do that will answer the question you are studying before diving into the literature.

Perhaps you will also come across experiments in the literature that are often done on RNA and give you a lot of data experimentally - perhaps you could model that system, or make an approximation of it.

Finally, you cannot use all types of experimental data to compare to MD. So don't be afraid to also include quantum and course grain techniques if they will be of more use to your problem.

If you are not in a group that work with MD, I would strongly suggest doing a research visit to a group that use the software you decide to use. It is very helpful to have people around you who face the same problems as you do.

Wishing you all the best!

>MD is hugely dependent on the starting structure you use and at the all-atom level is very poor at crossing from one minimum to another so make sure you have some evidence for your starting structure.

Hear, hear. Granted, you can use advanced techniques like REMD or metadynamics for better sampling, but you want to understand MD quite well before doing that (and you need plenty of computational resources).

you might check out lammps since it has GPU support. i've used it for condensed matter simulations and got huge speed ups vs CPUs. Also, a gpu is significantly cheaper than a ton of cpus.

RNA simulations are sensitive to the starting conditions - however you can model it with various softwares (I think people have mentioned some already). In addition, with RNA simulations, due to the high charge density, you need to be careful with charge placement and also with the solvation of the RNA and the ions (I already noticed someone has attached an Eric Westhof paper - other people to look up are Zan Schulten and Beveridge). If you are more specific about what you want to do with the simulations and what problems you are facing, maybe people can be of more help.

Hi, after having followed all the reported suggestions, in particular those regarding the understanding of the chemistry related to your question and of methods reported by litterature, you might have a look to the NAB code within the amber package, which has been designed for building nucleic acid sequences.

If your question is predicting conformations, as Massimo said, I would also suggest the use of highly efficient sampling techniques such as REMD (provided that you have the necessary computational power), followed by a cluster analysis of resulting trajectories at a given temperature (see REMD tutorial on the amber site).

Regarding the dependence of results on starting conditions -- it shouldn't, you MUST be sure that the initial, equilibration step has passed, otherwise, what you get at the end has no scientific value. Sorry for saying it in a direct manner.

Thanks all very much for your helpful replies. I really appreciate it. I come from doing biochemistry with RNA in the lab. I came across this interesting RNA hairpin (20 nt), which seems to have a repressor role in the process I'm studying. My current boss suggested this new project, which seemed very cool at the time. His lab (where I am in now) used to have many modeling people, but they have all left the lab. We have a GPU and all the programs I need, but I lack experience completely, and people around to ask. Well, so my boss' idea of the project was to model the folding of this 20-nt hairpin to understand some critical steps/position for its formation, with the idea of using this model later in a virtual chemical screening. I don't know if that is enough information really, but it's all I know at this stage. Before I moved here, I did some single mutagenesis in different positions of this RNA hairpin and got useful information on their functional impact in cell culture. Maybe that'll give me a clue as to what positions to focus on etc, but at the moment things are not much more defined. I'm excited about the project, but a little scared too. And I needed to hear what other experts thought. I have definitely not ruled out a short visit to another lab, and have actually spotted one not too far.Thanks everyone again for your ideas and help!

I would suggest you to study the code of the programs you have and read books in parallel. There is a bunch of good books on simulations. First step I highly recommend: take a look at a part in program where potentials are and try to read in the literature on what do they mean, limitations, problems and advances. Second, find out, whether it is molecular dynamics MD (there are several versions of it, be aware) or Monte-Carlo MC (again many versions of it). Third, go to the part that establishes the polymer model -- is it beads on a spring, or whatever. Then try to reproduce the previous results, obtained with the same code. You will get the hint, what is crucial to get right results. Run the real, to-be-published simulation only when you know what every line of the code does. When doing in vitro experiments, you have to exactly know, what compounds are present inside your experimental cell or setup, right? Same here, it is not a magic box, but rather a model. And keep thinking that it gives you what you have put in. The believe is that if every step (polymer model, potentials, MD or MC motor, etc...) is done in a self-consistent, not contradictory way, the result would be related to the real life (what you see in vitro). If you have practice with real life experiment, you could probably imagine, how crazy can things go, for example, if there is dust in the water, or not well-purified sample, or something else. Same about doing simulations. Any approximation and model has its limitations and the range of applicability. You should get clear, what is going on inside the program. And good luck! That's a pity none to ask in the lab. After some reading and trying, you will find out, that you have tons of questions. To avoid re-inventing bicycle, if there is a possibility go to another lab, who is doing simulations like yours, say, for a couple of weeks. That would be extremely useful too. All the best and keep asking.

I am stepping in very late in this discussion but . . . REMD and metadynamics were mentioned previously. I would echo what other people have said about understanding the question you are trying to ask about this RNA. For example, whether you are interested in the most likely structure, or the mechanism by which it folds, or the rate at which it folds, etc. It is also crucial to know your software, your model, and statistical thermodynamics inside and out if you want to be confident that you are "doing it right." However, once you know the question you want to ask and understand all the tools you are applying to get there, you should investigate advanced sampling techniques. I, for one, see little value in long, brute force simulations when one can perform metadynamics and directly uncover the underlying free energy surface for a useful order parameter - the brute force simulation will converge to the same answer in the infinite time limit but, if the barriers separating important states are high, the brute force simulation may not converge in a useful time frame. Or if you are interested in folding pathways, while brute force simulations may eventually teach you something useful, enhanced sampling techniques like Transition Path Sampling or Forward Flux Sampling may get you there much more quickly, and in a manner you can defend in a paper as being rigorous. There are many other enhanced sampling methods and all have strengths and weaknesses but ALL tend to be more useful in answering specific questions than simply running long simulations. So learn your code, your model, understand the theory, and formulate the specific question you wish to answer... then see what awesome algorithms, methods, tools exist to answer that specific question more rigorously than simply running the system for a long time and seeing what happens.

To Gordon: Exactly! And moreover, you never know if your "brute force" model has anything to do was the real system, or not. Too complicated and uncontrollable. As to sampling -- if not interested in crystalline structures, Wang-Landau microcanonical sampling (Monte Carlo) is a very powerful technique, which will give you DOS as an output. So that you can just multiply it by Boltzman factor to get canonical distribution, and hence, the complete thermodynamics of your system.

Hi Amparo,

All very good suggestions about sampling techniques and limitations!

I would add to take a look at the limitations of the force field you will choose to define your RNA fragment. Your model can be only as accurate, thus valuable, as the description the force field gives of it. Make sure you have good enough confidence in your chosen force field, so that you won't spend the whole year sampling thoroughly the wrong conformational space.

For instance, I do know for sure that the sugar parameters in DNA and RNA commonly used force field are particularly tricky. First thing I'd do is to go through the literature about RNA/DNA modelling with a fine tooth comb.

Best of luck!

To Elisa: one comment, if I may. RNA in its native form (with secondary structures intact) is better described as a branched polymer. Also, its properties as polyelectrolite are VERY different from DNA. Additionally, the persistence lengths of two systems (RNA vs DNA) are different too. So that the force fields MUST be unavoidably different.

Good luck!

Artem

Artem, As far as I know , most parameters are the same for both DNA and RNA in the most popular nucelic acids force fields (Amber,. CHARMM, OPLS). Some atom types for uracil and ribose are obviously exclusively for RNA, the other atom types are the same, unfortunately you may add, but they are...

If we forget about the details of simulation packages for a moment, and look onto the systems under study, we immediately see that in both ordered and disordered states RNA and DNA are very different. Both ordered (3D structure for RNA, helix for DNA) and disordered (single strand coil for RNA vs two inter-wound strands for DNA) conformations differ. There are 2 negative charges, sitting in front of each other and repelling the strands on DNA, while in RNA the ratio of secondary structured repeated units is much smaller, hence, such a repulsion is weaker. There is also a very strong dependence on the ion environment, since both systems are polyelectrolites. Another point -- persistence lengths are differ significantly.

To summarize -- systems are very different at all levels, so that force-fields cannot be the same.

use iFoldRNA for modelling of your RNA .

Hi!

In my lab we regulary perform molecular dynamic simulation of RNA. I recommend you to use RNAComposer online software to construct the 3D structure of your RNA (http://rnacomposer.cs.put.poznan.pl). Then I use gromacs software and CHARMM27 v 2.0 force field that works well.

It is difficult because there are so little papers about it but it is possible. You have to be careful especially in the 3D structure construction and compare with similar DNA which has been reported.