Chapman-Kolmogorov validation

[jchodera]

We should be using the Chapman-Kolmogorov validation on our final MSMs as another check of self-consistency.

[maxentile]

Good call! I had mentioned that in the readme but didn't include it yet. Will add few lines for this in my next pull request.

We'll need to specify m groups of states of interest here. Shall we just use whatever macrostate definitions we settle on here?

[jchodera]

It might be good to check out the best practices MSM review from Frank and I to see what the specific recommendations is. The number of trajectories cisiting individual microstates is usually so low that the correlation functions are very noisy, so I believe they recommend using random collections of macrostates whose total probability or counts are above some threshold.

[maxentile]

Thanks, John!

Sorry, I'm having trouble finding your reference that recommends to use random collections of macrostates-- could you please point it out to me?

• In section IV.F. of MSMs of molecular kinetics: generation and validation (2011) (and Chapter 4.8 of the MSM book), you used the top 6 metastable states identified by PCCA.
• In MSMs of biomolecular conformational dynamics (2014), the CK test doesn't appear to be explicitly mentioned.
• In Automatic discovery of metastable states (2007), 3 states were monitored in the CK test (Figure 6 + section IV.B.2), but I didn't see where the method used to select those states was described.

[jchodera]

In section IV.F. of MSMs of molecular kinetics: generation and validation (2011) http://www.diss.fu-berlin.de/docs/servlets/MCRFileNodeServlet/FUDOCS_derivate_000000005544/1.3565032.pdf (and Chapter 4.8 of the MSM book http://www.springer.com/us/book/9789400776050), you used the top 6 metastable states identified by PCCA.

I think this was the review I was thinking of!