Atlassian uses cookies to improve your browsing experience, perform analytics and research, and conduct advertising. Accept all cookies to indicate that you agree to our use of cookies on your device.
Atlassian uses cookies to improve your browsing experience, perform analytics and research, and conduct advertising. Accept all cookies to indicate that you agree to our use of cookies on your device. Atlassian cookies and tracking notice, (opens new window)
MT – what’s the current status of the sage release?
DM – Not ready to release release candidate - Inferior in some metrics, even if equivalent/superior to others. Some options for what to release, including 2.2.1+NAGL in place of ToolkitAM1BCC
JW – what do we think is the timeline on the protein force field in the best-case scenario (where the next experiment passes all of Chapin’s benchmarks)
DM – Not a super clear sense of timeline/progression in that case.
JE – IMO, we should move as quickly as possible toward a release in that case.
JAC – Can’t recall if CC is using NAGL charges
JW – CC is using them, not via NAGLCharges tag that was implemented last week, but via ChargeIncrementModel backdoor.
DM – Kinda a different case here than normal, since goal of talk is to inform ad board about cool work that’s happening… in this case the effect would be sweeping dataset cleaning details under the rug and not showing training curves.
FC – Audience I wrote it for was for people in management-y position at companies who are supporting OpenFF. I wanted them to know that there was a new fitting pipeline that was potentially faster and leading to improvements. Could have been clearer about that message.
JE – Right, first and last slide should say that.
FC – Could cut down a lot of detail in these.
JE – How would you cut this down to 4 slides?
FC – (see recording ~42 mins)
JAC – I kinda think some of the ones you were thinking about cutting were valuable - Some of the separate slides here were basically animations, and those are good ways to show movement/keep attention.
DM – Agree.
MT – A reflection is that, if we wanted to apply principles from book to our meetings, the key principle would be to move a lot of the middle data slides into kinda SI/hidden post-conclusion slides. Eg. every plot from YAMMBS could go into supplemental slides. I think most YAMMBS slides I see communicate “these results are nearly identical to previous results” and it takes a lot of time to scrutinize them and get that point.
DM – Also important to understand what people want to get out of their presentation - Am I here to help with data analysis, or just get a summary.
JAC – Right, sometimes I don’t know what I’m gonna get/where in the meeting I’ll get what I want
JW – Would be interesting to have something like a running timeline for meetings where there’s an estimate of when agenda items will come up, and when feedback/interruptions are welcome.
JE – Shout out to FC for the great ad board update.
JW – MT has been extra on-top of things in the past couple of iterations. Working on lots of different repos, closing issues before I noticed they were open, and also pushing forward collaborative work with OpenFE.
MT – Credit is due to Irfan for the vast majority of pontibus.
JAC – Shout out to CBayly for starting a great conversation at ACS that will probably bear fruit longer term.
LW – Jen went to ACS and came back with lots of suggestions for collaborations.
Q&A
CC – it seems like Smee is absolutely crushing it on valence fits, so do we have criteria or a decision process for when / how we will switch from ForceBalance?
LW – Finlay’s work has already crossed the QM quality threshhold. Currently, the team is doing a cross-training “camp” where three out of four team members are working on fitting experiments with Smee. The moment when we will have fully switched over is when we release a force field that was fit with smee. We are not planning to do any new fits relying solely on ForceBalance after the current 2.3.0 work. But I don’t have a timeline now for when we will release a fully smee-fit force field.
Of particular interest to CC is that JM is trying out fitting (using smee) with and without the 4-mer dataset.