search

markovmodel / deeptime

Deep learning meets molecular dynamics.
GNU Lesser General Public License v3.0
175 stars 39 forks source link

What does a visual picture mean? #27

Closed zhouhao-learning closed 6 years ago

zhouhao-learning commented 6 years ago

Here are four pictures that I got by executing example / Alanine_dipeptide_multiple_files, but I can't understand what they mean, so can I explain them in a nutshell? The picture is shown below: 1 default default default

Thank you very much!

amardt commented 6 years ago

Hey,

so I really recommend to have a look in the paper, there these kinds of figures and the theory behind it are explained. But in a nutshell:

  1. Its a state separation plot. Each color represents a state (where the output node with the highest activation wins). For Ala-Dipeptide with 4 output nodes you hope that the model sets 2 states on the left and 2 on the rigth side of this plot. (So your model is not optimal)
  2. Its a similar plot but this time for each output node (state) you plot the activation for each input.

After you have a Markov State Model you can test it with the the implied timescales (ITS) and the Chapman-Kolmogorov Test (CK-Test).

  1. ITS plot, where you get for each process a line. The processes you find (the ones not in the gey area) should be converged.
  2. CK-test. You check if prediciton with a model which is estimated on tau agrees with a model estimated on n*tau. Meaning: The red and the blue line should be on top of each other.

Again: The paper states these stuff in far more detail. Best Andreas

zhouhao-learning commented 6 years ago

@amardt Ok, thank you very much, mainly because I don't have knowledge of molecular dynamics, so I don't understand your paper.