Rovibrational Entropy

[ioanaapapa]

Rovibrational entropy does not match results obtained using the old code (version 0.3 with scaling) or results from the host-guest SAMPL8 paper (https://link.springer.com/article/10.1007/s10822-021-00406-5) for G1. The trend is similar for the host and other guests for new results, new code compared to paper results. There is a much better match for biotin for the new AMBER trajectories but we see some slight differences for the united atom rovibrational term, which are probably because of the axis changes. (will investigate further about the axes)
All data for G1 is for 1001 frames and all data for biotin is for 1000 frames.
For small organic molecules - biotin and G1 from the host-guest systems:

For the host and guest molecules, at the residue level, rovibrational entropies obtained using the two versions of the code are similar but always much lower than the value in the paper, I don't understand why and we can discuss this with Richard at the next CodeEntropy catch-up. At the united atom level, the result obtained using the old code are close to the SAMPL paper results,as expected but results obtained using the current release are very small. [lo]

[harryswift01]

Thanks for letting us know about this @ioanaapapa. Just so we're able to pin this issue down, which version of CodeEntropy are you currently using?

[ioanaapapa]

I was using 1.0.4. The one referred to as old code is the last 0.3 stable release.

[ioanaapapa]

Updated values for biotin to include older gromacs trajectories analysed with 1.0.4. Differences only appear at the united atom level. Same small differences for the transvibrational term for the different trajectories but not too significant I believe; there is no axis change for this term as the principal axes of the molecule are relevant here.
More significant differences for the rovibrational term, likely due to change in axes ( however, not as significant as for the guest). However, there doesn't seem to be a consistent trend?
Everything here is obtained from 3 replicates, 1001 frames, 50ns for consistency (however, time is not very significant for biotin as there isn't any significant change in values for a small molecule).

[ioanaapapa]

In the alternative axes version, united atom level rotation axes are based on covalent bonds with the hydrogens, which should be more similar to the way the were chosen in the old implementation. It doesn't change much at the united atom level.

Differences between the two codes, if changing the force partitioning factor for G1 (note: this would also change transvibrational entropy at the residue level but that term matches previous results and results obtained with different code versions):

Torques for 1 frame [note: in the current code, we multiply forces by the scaling factor, then get the torques, then construct the matrix; meanwhile, in the old version, we got the torques, constructed the matrix then scaled within quadrants of the matrix]:

weighted_torques_G1_new.txt

weighted_torques_G1_old.txt

At the residue level, the axis choice is the same (principal axes) so torques should be the same. They are already different before the inertia weighting.

[ioanaapapa]

Issues related to axes:

• axis in the moment of inertia are swapped compared to the torque axes
• axes at intramolecular levels should be based on bonding for the moment of inertia, not principal axes
• transvibrational axes should match rovibrational axes at upper level