Re: Improving MDFF results

From: Tristan Croll (tristan.croll_at_qut.edu.au)
Date: Tue Oct 21 2014 - 14:22:31 CDT

Hi Kevin,

A gscale value of 0.5 is still a very high starting point. I'd be starting much lower - 0.05 or 0.1.

Cheers,

Tristan

Tristan Croll
Lecturer
Faculty of Health
School of Biomedical Sciences
Institute of Health and Biomedical Engineering
Queensland University of Technology
60 Musk Ave
Kelvin Grove QLD 4059 Australia
+61 7 3138 6443

This email and its attachments (if any) contain confidential information intended for use by the addressee and may be privileged. We do not waive any confidentiality, privilege or copyright associated with the email or the attachments. If you are not the intended addressee, you must not use, transmit, disclose or copy the email or any attachments. If you receive this email by mistake, please notify the sender immediately and delete the original email.

On 22 Oct 2014, at 1:56 am, "Kevin Chun Chan" <mcfc1301_at_gmail.com<mailto:mcfc1301_at_gmail.com>> wrote:

Thanks Tristan for very clear reply.

So does it mean that I should go back to my 0.5 gscale simulation results and continue to run a longer time? I admit that I was in a rush for a better CCC so I increased the gscale. Before increasing the gscale, the RMSD went flat at about 2ns (also the CCC) and I waited for another 5ns so I thought it has already converged.

Now I understand that CCC is a crude measure. But also as what I am interested in is the very detailed contacts between two domains of the proteins, even if it looks like fitting the density map, how could I say that? Should I run another few ns with 0.5 gscale to see any further improvement though I don't think there will be.

Cheers,
Kevin

On 21 Oct, 2014, at 13:36, Tristan Croll <tristan.croll_at_qut.edu.au<mailto:tristan.croll_at_qut.edu.au>> wrote:

Hi Kevin,

If you have a system that is poorly fitted to the map, increasing gscale is the last thing you want to do – it will only induce over-fitting. In general, what you want to do is start with a very low value so that the normal dynamics of your protein dominate, and it can escape from poorly-fitting configurations. Only once it’s converged at a low coupling constant should you start to step up the gscale value.

In the end, CCC alone is a fairly crude measure. Nothing compensates for direct inspection of the structure against the map. Look it over, identify the portions that are poorly fitting, and then decide what to do about it. If the problem is obvious (e.g. a domain “bridging” across two distinct density blobs) you might for example consider a short TMD simulation on top of MDFF to bring it back on track.

Cheers,

Tristan

From: owner-namd-l_at_ks.uiuc.edu<mailto:owner-namd-l_at_ks.uiuc.edu> [mailto:owner-namd-l_at_ks.uiuc.edu] On Behalf Of Kevin C Chan
Sent: Tuesday, 21 October 2014 1:16 PM
To: namd-l_at_ks.uiuc.edu<mailto:namd-l_at_ks.uiuc.edu>
Subject: namd-l: Improving MDFF results

Dear NAMD users,

I am currently doing MDFF simulations to investigate some conformations of a protein. The simulation was done in a water box and with a low-resolution electron density map (14A). The system is quite large (>1M atoms) so I have just run for several ns.

By analysing RMSD and CCC of the simulation results, the system seems to converge in the first 2 ns. However the CCC of the system relative to the density map started at 0.60 and converged at 0.65 which was sadly not much improvement. I have step-wisely increased the g-scale value from 0.5 to 3.0 and the CCC further improves to 0.69. After that, I am not able to increase the g-scale value as the simulation will be terminated due to “atom moving too fast” (I guess it was due to the increasing magnitude of the additional potential right?). I understand that such a low-resolution density map will definitely limit the maximum CCC that I can have but is there any other ways to improve the MDFF results?

I will greatly appreciate that if anyone could share your experience on this.

Thanks in advance.

Cheers,

Kevin
ukevi_at_gmx.hk<mailto:ukevi_at_gmx.hk>

This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:21:19 CST