Re: [AMBER] PCA multiple trajectories for mutants and wild type

From: Wesley Michael Botello-Smith <>
Date: Wed, 11 Jul 2018 11:07:36 -0700

One option would be to use RMSD clustering in CPPTRAJ. Assuming you are
starting from the same crystal structure, you could align each trajectory
to the crystal structure and run kmeans to get an estimate of the different
states that each trajectory explores.
Due to the degeneracy in an RMSD measurement, this is not a perfect metric,
but it may give a decent lower bound estimate for how many different states
each trajectory explored.

For more detailed investigations you could use the perres option to get the
per-residue RMSD in each trajectory. I like to import these into R or
gnuplot to produce a 'heatmap' plot to get idea of what portions of the
protein have changed at each step. E.g. the x axis is time, y axis is
residue number and the color maps to RMSD. While there is still a good
degree of degeneracy involved, it is more 'finger print' like for various
Wavelet analysis (WAFEX) is another option as well.

Additionally, you could employ clustering techniques on the per-residue
RMSD / wavelet data to attemp to identify individual states.

Another option would be to compute phi / psi angles (ramachandran). Using a
heatmap approach you could again get a sort of 'fingerprint' of backbone
conformation at each step.
Also, in addition to using PC on cartesian coordinates, consider applying
it to the phi / psi backbone angles as well. These act as internal

If you have sufficient trajectory lengths, you may also consider using
Markov state modeling techniques... PC is certainly one option there,
although TICA seem to be more favored for Markov state modeling.

On Wed, Jul 11, 2018 at 4:48 AM, Stejskal, Lenka <> wrote:

> Dear Amber users,
> I am trying to compare multiple trajectories of either wild type and a few
> mutant of the protein to compare differences in their dynamics. I have
> several ideas how to do this and any experience/ideas would be appreciated.
> I create the matrix based on all trajectories and project the 1st and 2nd
> PC onto individual trajectories to see a difference in the occupied phase
> space. Alternatively, as I am trying to compare the mutants to the wild
> type, I could create the matrix based just on the wild type and project
> trajectories of the mutants on these eigenvectors. I believe this might be
> a better measure.
> I am however making an assumption that the motion is comparable. Could
> this be an issue? Is there any way I can confirm this?
> I will also calculate the eigenvectors for each trajectory individually.
> Is there a way to compare how similar these are?
> Thank you very much for your help!
> Lenka
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Wed Jul 11 2018 - 11:30:03 PDT
Custom Search