Two Picoseconds Simulation of Protein

[porchrat]

+-135000 atoms in a 10x10x15 box including POPC (about 130), Cl, Na, K, protein and GABA.
I used my desktop PC. Xeon 1230v2 with a 660ti running on Ubuntu 12.04. I got about 8ns/day (Gromacs 4.6.3 and Verlet cut-off) so it ran for a little more than a day.

I used the AMBER forcefield for this simulation. VMD was used to insert the protein in the lipid bilayer. Afterwards the the system was minimized and equilibrated (pressure and temperature) before doing the simulation.

The crystal structure for the GABA receptor does not exist so I used the Bergmann homology model (published in PloS ONE). The problem with this is that the intracellular loop domain is missing so I am now buzy trying to make good homology models of these domains. So far I-Tasser seem to do a reasonable job for this.

Working with incomplete data is rough. These experimentalists need to stop being so lazy and catch up with the molecular dynamics boys .

Not having parameters sucks too. Extending force fields looks like one hell of a mess I just wouldn't want to dabble in. I've had supervisors tell me some horror stories XD. At least the protein parameter sets appear relatively well fleshed out. I can't remember seeing people missing parameters for whole amino acids. Could be wrong though, don't really have much experience there.

Yea I can see you're using VMD now. I recognise the axis label. It has a hideous interface but it does a wonderful job. The addition of a Tcl console really helps.

I'm amazed you got so many nanoseconds on a system so large out of a single day of processing. You running all 4 cores on that Xeon together? Have you tried running on a single core and if so what difference did you see (i.e. is there a noticeable overhead when you switch to multiple cores)? Also don't you mean you ran it for slightly less than a day?... 5ns sim... 8ns per day? Math seems off.

By all means keep the videos coming I really enjoy seeing them.

 

[BaSuP]

Working with incomplete data is rough. These experimentalists need to stop being so lazy and catch up with the molecular dynamics boys .

Not having parameters sucks too. Extending force fields looks like one hell of a mess I just wouldn't want to dabble in. I've had supervisors tell me some horror stories XD. At least the protein parameter sets appear relatively well fleshed out. I can't remember seeing people missing parameters for whole amino acids. Could be wrong though, don't really have much experience there.

It took some time to combine AMBER protein and AMBER lipid force fields into one one set. I am just glad it looks like it is working, or so it seems anyway.

Yea I can see you're using VMD now. I recognise the axis label. It has a hideous interface but it does a wonderful job. The addition of a Tcl console really helps.

I am slowly warming up to vmd .

I'm amazed you got so many nanoseconds on a system so large out of a single day of processing. You running all 4 cores on that Xeon together? Have you tried running on a single core and if so what difference did you see (i.e. is there a noticeable overhead when you switch to multiple cores)?

The GPU is responsible for the non-bonded calculations (e.g.water) and speeds it up considerably (up to 5x, maybe even more). The CPU uses all 8 thread (4 cores with hyper-threading) and does the rest.

Also don't you mean you ran it for slightly less than a day?... 5ns sim... 8ns per day? Math seems off.

By all means keep the videos coming I really enjoy seeing them.

Yeah, my mistake. I got confused. I saw the run file that there was enough steps for 10ns. I will have to fix the youtube link .

 

[BaSuP]

A protein-ligand simulation.
[video=youtube;Ff87lPkz1L4]http://www.youtube.com/watch?v=Ff87lPkz1L4[/video]

A force was applied to the ligand in order to pull it from the protein it is bound. From this, various other simulation were run and from there the Gibbs free binding energy was calculated:

 

[porchrat]

It took some time to combine AMBER protein and AMBER lipid force fields into one one set. I am just glad it looks like it is working, or so it seems anyway.

Well at least you don't have atoms zipping about or anything. Looks nice and stable from the trajectory, for what that is worth

I am slowly warming up to vmd .

I have been using it for a while now and I still hate the interface. One of the ugliest GUIs I've seen in a long time. Still I can't deny that it has a lot of very nice features and you get used to the interface after a while.

Have you tried writing up any Tcl scripts? You can do some nice stuff with that language. Wrote a little script a while back that extracts a list of frames fitting a particular description (certain angles or bond lengths or something) from a trajectory and creates a new smaller trajectory from them that is easier to work with. Maybe not that big a deal on smaller sims but when you have sims above 1000ns with millions of frames then I've found being able to do that sort of thing really speeds analysis along.

The GPU is responsible for the non-bonded calculations (e.g.water) and speeds it up considerably (up to 5x, maybe even more). The CPU uses all 8 thread (4 cores with hyper-threading) and does the rest.

Ah that would explain it. I've only ever run on CPUs in clusters and even then it takes forever. I've been wanting to try out a GPU based MD sim. Would be nice not to have to wait so long for results.

 

[BaSuP]

I haven't tried using the tcl scripts yet. I will ask for some tips once I get these GABA systems running for microseconds. Can't wait untill the CHPC finish setup their K20 clusters. Anyway, here is a 1ns simulation of the human Glycine receptor α1 in a POPE membrane. Took half a day from setup to finish. It's nice/interesting to see those waters move away from the lipids like that.
[video=youtube;JnGBod1QRLU]http://www.youtube.com/watch?v=JnGBod1QRLU&feature=youtu.be[/video]

 

[porchrat]

I haven't tried using the tcl scripts yet. I will ask for some tips once I get these GABA systems running for microseconds. Can't wait untill the CHPC finish setup their K20 clusters. Anyway, here is a 1ns simulation of the human Glycine receptor α1 in a POPE membrane. Took half a day from setup to finish. It's nice/interesting to see those waters move away from the lipids like that.

I wrote a few simple scripts to organise the data for more streamlined analysis but I'd hardly call myself an expert on Tcl. If you need help getting started I might be able to help but I doubt I'd be able to help with anything too advanced.

I did most of my analysis code in Java. VMD allows you to output labels and their frame numbers as plain text files (tab separated). I just used those text files as inputs and processed those. The code to do that is rather simple.

I only bothered to look at Tcl when I encountered a program called CatDCD that did what I needed and could be easily used through Tcl. If you have any experience in coding then Tcl is pretty easy to pick up. Having seen how easy it is to use now I wish I had started using it sooner.