- Contemporary messages sorted: [ by date ] [ by thread ] [ by subject ] [ by author ] [ by messages with attachments ]

From: Daniel Roe <daniel.r.roe.gmail.com>

Date: Thu, 26 Apr 2018 13:40:36 -0400

Hi,

Note that wavelet analysis is still fairly new in cpptraj, so it is

still fairly uncharted territory - there's going to be a lot of

experimentation involved. Your instinct to start with a smaller number

of frames is a good one!

On Tue, Apr 24, 2018 at 1:42 PM, Moataz Noureddine

<moataz.noureddine.gmail.com> wrote:

*>
*

*> parm structure_stripped.parm7
*

*> trajin combined_stripped_aligned100ns.nc
*

*> center :1-160 mass origin
*

*> image origin center
*

*> rms first *
*

You may want to try just using 'autoimage' instead of image/center.

Also, if you're going to be doing several rounds of wavelet analysis

it makes sense to create a stripped, imaged, and rms-fit trajectory

(i.e. combined_stripped_aligned100ns.nc should be stripped, imaged,

and fit) so you don't have to constantly image/fit.

*>
*

*> wavelet nb 54 s0 0.01 ds 0.25 correction 1.01 chival 1.6094 type morlet out
*

*> wavelet_stripped100ns.gnu usemap
*

*>
*

*> First, do those seem like reasonable parameters for our analysis?
*

The value of s0 seems too small here to me. The parameters for wavelet

analysis can be a little tricky (as you're finding out). Essentially,

the time scales you get out depend on the time scale of the frames

going in. If the input frames are 1 ps apart, an initial scale value

(s0) of 1 means the wavelet will be 1 ps. The smallest value you

should choose is 2*delta, where delta is the time between your

snapshots. So if your snapshots are 1 ps apart, you should choose an

initial scale value of 2. The number of scaling values (nb) will

determine the max scaling value (s0*2^nb*ds), which in turn determines

the max length of "significant" motion that will be detected. In your

case, your scales range from 0.01 to about 116 - assuming your

trajectory save frequency is on the order of ps that would explain why

the resulting wavelet plots look so flat. You'll probably want to run

the wavelet analysis several times with a higher s0 and increasing

values of nb to reveal different timescale motions.

If you haven't already, you may want to check out this paper:

https://pubs.acs.org/doi/abs/10.1021/acs.jcim.5b00727, which discusses

the implementation of wavelet analysis in cpptraj, and some of the

references therein (e.g.

https://www.worldscientific.com/doi/abs/10.1142/S0219691312500403)

which discuss application of wavelet analysis to biological systems.

Also note that the wavelet analysis code is OpenMP-parallelized, so

you'll get a speedup using cpptraj.OMP (up to the number of real cores

on your system, no hyperthreading).

Hope this helps,

-Dan

Date: Thu, 26 Apr 2018 13:40:36 -0400

Hi,

Note that wavelet analysis is still fairly new in cpptraj, so it is

still fairly uncharted territory - there's going to be a lot of

experimentation involved. Your instinct to start with a smaller number

of frames is a good one!

On Tue, Apr 24, 2018 at 1:42 PM, Moataz Noureddine

<moataz.noureddine.gmail.com> wrote:

You may want to try just using 'autoimage' instead of image/center.

Also, if you're going to be doing several rounds of wavelet analysis

it makes sense to create a stripped, imaged, and rms-fit trajectory

(i.e. combined_stripped_aligned100ns.nc should be stripped, imaged,

and fit) so you don't have to constantly image/fit.

The value of s0 seems too small here to me. The parameters for wavelet

analysis can be a little tricky (as you're finding out). Essentially,

the time scales you get out depend on the time scale of the frames

going in. If the input frames are 1 ps apart, an initial scale value

(s0) of 1 means the wavelet will be 1 ps. The smallest value you

should choose is 2*delta, where delta is the time between your

snapshots. So if your snapshots are 1 ps apart, you should choose an

initial scale value of 2. The number of scaling values (nb) will

determine the max scaling value (s0*2^nb*ds), which in turn determines

the max length of "significant" motion that will be detected. In your

case, your scales range from 0.01 to about 116 - assuming your

trajectory save frequency is on the order of ps that would explain why

the resulting wavelet plots look so flat. You'll probably want to run

the wavelet analysis several times with a higher s0 and increasing

values of nb to reveal different timescale motions.

If you haven't already, you may want to check out this paper:

https://pubs.acs.org/doi/abs/10.1021/acs.jcim.5b00727, which discusses

the implementation of wavelet analysis in cpptraj, and some of the

references therein (e.g.

https://www.worldscientific.com/doi/abs/10.1142/S0219691312500403)

which discuss application of wavelet analysis to biological systems.

Also note that the wavelet analysis code is OpenMP-parallelized, so

you'll get a speedup using cpptraj.OMP (up to the number of real cores

on your system, no hyperthreading).

Hope this helps,

-Dan

-- ------------------------- Daniel R. Roe Laboratory of Computational Biology National Institutes of Health, NHLBI 5635 Fishers Ln, Rm T900 Rockville MD, 20852 https://www.lobos.nih.gov/lcb _______________________________________________ AMBER mailing list AMBER.ambermd.org http://lists.ambermd.org/mailman/listinfo/amberReceived on Thu Apr 26 2018 - 11:00:02 PDT

Custom Search