Feat/cyclic offset #526
Open
Feat/cyclic offset #526
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Would be a great merge. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
The purpose of the PR is to implement the cyclic offset positional encoding in OpenFold. Currently, the residue relative positions are computed linearly, meaning that the sequence is assumed to be a linear peptide. However, this relative positioning does not work well in cyclic peptides. As such, this PR applies the cyclical relative position encoding to allow for better prediction of cyclic species as first reported in the AfCycDesign preprint.
We are showing the result of applying the cyclic offset on the cyclic peptides tested by the AfCycDesign authors (see figure below). We are comparing the RSMD achieved using the standard linear offset with the RSMD achieved using the cyclic offset. The figure shows that for the majority of the peptides the cyclic offset achieved better alignment with the ground truth compared to the linear offset.
We note that we are using the unrelaxed version of the OpenFold output as we have noticed that the relaxed version does not preserve the cyclization bond.