Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update the way we do retrospective compound analyses #174

Open
jchodera opened this issue Feb 7, 2022 · 0 comments
Open

Update the way we do retrospective compound analyses #174

jchodera opened this issue Feb 7, 2022 · 0 comments
Labels
bug Something isn't working

Comments

@jchodera
Copy link
Member

jchodera commented Feb 7, 2022

Our current retrospective analysis will now shift back to comparing the accuracy of individual microstate transformations with the experimental data differences between compounds, now that we have more accurate microstate stereochemistry assignments.

When we return to benchmarking the accuracy of compounds, we currently use the estimate derived from running DiffNet with all available experimental data. This is not sensible for assessing the accuracy of predictions for compounds whose experimental data is available, since the data for that compound will contaminate the benchmark.

Instead, we should consider a few different variants:

  1. We can use the DiffNet solution where no experimental data has yet been included, and all computed DeltaGs are shifted by a constant to minimize the RMSE
  2. We can use the DiffNet solution where a single experimental measurement has been included and no constant is added to the computed DeltaGs
  3. We can use the DiffNet solution where all other experimental measurements (except the compound in question) have been included, and no constant is added to the computed DeltaGs.
@jchodera jchodera added the bug Something isn't working label Feb 7, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working
Projects
None yet
Development

No branches or pull requests

1 participant