Individual updates | DN Working a lot with OpenFE. It needs more immediate attention since OpenFF is already established/more mature. OpenFE needs more protocols/process to help understand how decisions are made. Followed up on water models question with JW at last lead team meeting. Reaffirmed our goals to keep org details away from individual contributors and let them make cool stuff. Lots of ad board comms - they’re asking for details on everything. Takes a lot of time, hard because there are a lot of details to communicate and english is not my first language.
CC Still running parameter fit (null model, no new torsions, but yes librarycharges) for proteins in ForceBalance on TSCC. Objective function changes are becoming small (single digits) in each iteration. So I think we’ll have the first generation of the protein FF soon. We expect null model to be comparable to amber99. Worked on SMIRKS for protein-specific torsions. My first attempts were having a lot of overlap between rpotein and small molecules. There are a lot of edge case where small molecules include a ring that looks like (for example) fluorinated proline. But I think it’s reasonable for those to get protein FF parameters due to the similarity. PB – It would be good to check the cis/transness of the proline in the small molecule to ensure that it doesn’t interfere with the protein parameter training. CC – In this case, the small molecule would get the proline backbone parameter, but not the sidechain parameter. PB – One tricky thing here would be if the surrounding context in the small molecule leads the proline-like group to be more planar than in a protein context. (eg there’s a larger resonant system past the amide bond in the small molecule, that could require a different parameter than the protein one)
Worked on prepping benchmark suite for proteins. Unfortunately a lot fo the NMR ata is colelcted at pH 2. This would have us expect charged N termini and uncharged C termini. Unfortunately ff14sb doesn’t have charge parameters for neutral C termini. It looks like other work in this area just uses a charged C terminus, but that’s not going the be the actual state at pH 2. I’m trying to track down the most appropriate parameters for use for this. Emailed Carlos Simmerling. JW – Maybe just add an NME cap if there are no uncharged C terminus parameters? Intuitively, the steric difference might be “less bad” than the charge difference. CC – Seems like there’s no great answer here unless we get one directly from the AMBER devs. I’ll poll the protein FF group to decide what to do if we don’t hear from Carlos.
PB Fitting related work with energy levels. Met with BSwope and LPWang, had a good discussion about LPW's earlier work with Amber-FB15 (LPW saw cases where MM local minima were lower than the QM, so he targetedly optimized to those badly described minima and resolved the issue). Will try to use gradients as well in fits ala abinitio target. CC – So LPW generated new single point data in those poorly described minima? PB – LPW did MM minimizations, and then did QM single-points at the MM minima, and then optimized to those. CC – … PB – LPW mentioned that torsiondrive targets are derived form ab initio targets, so I assume it does compare relative energies.
Ken Takaba's qca PR. Looked into validation and submission fails on github actions because of conda resolving importlib_metadata to an older version (v1.5) and failing a line in toolkit's forcefield.py. Sage manuscript revisions, some updates and lot of latex formatting, will go out for author review today.
JW – Hosted final bespokefit workshop. JM did a great job and will be editing and uploading videos for remaining workshops. DN – In the future, feel free to have me do more manual stuff like video uploads. JW – Thanks. In this case I think JM is uniquely qualified to edit+upload, but I’ll tag you in to make the zenodo entries once those videos are up on youtube.
Focusing on cleanups/issue tracker reduction for the rest of the year. Working to update the rest of our stack for Toolkit 0.11 - There’s a branch of our fitting stack that depends on a new forcebalance release so I’m pushing for progress on that again.
|