Larger than expected RMSDs returned by webserver

[timt51]

Thanks for creating the folddisco web server!

When searching with the 1BCF.pdb motif as the query, I found that the reported hits generally look good, but have surprisingly high RMSDs. As expected, the server reports 1BCF, the original structure from which the motif came from, as a hit. However, it also reports that the RMSD to this hit is >8 (see image below). Looking at the reported alignment, it looks like the high RMSD is due to some of the residues being misaligned. Is this behavior expected?

Here are the links to the jobs:

https://search.foldseek.com/result/folddisco/l8zEJ2gU14SyUuEsB4ATIoptIDbgV8UMtNTFYA
https://search.foldseek.com/result/folddisco/ikvZ4e9SLfw1-yBsZKEs8LN67Y2akq5TLug6QQ

The jobs only differ in the order in which the residues are listed in the query pdb file.

[khb7840]

Thanks for checking Folddisco.
In residue matching step, Folddisco additionally matches residues based on C-α distances + AA pairs, ignoring sequence order; a few misordered hits can then skew the superposition and push the RMSD higher.
This step works fine to fill missing residues in small motifs, but for longer motifs, it can introduce false positive residues that are not part of the motif.
We have confirmed that for results with long motifs, IDF score is better in ranking than RMSD due to this false residue introduction.
We're aware of this issue and are working on improvements to the matching step.
Some ideas include:

1. TM-score normalized as an alternative metric
2. Alternative residue matching algorithms with less false positive residues
3. Improvement of superposition to handle motifs well

Thank you again for your report.
For now, we recommend using IDF score for ranking long motifs.
We'll keep you updated on our progress with improvements in residue matching and superposition.

[timt51]

Got it, thank you for the explanation!

[khb7840]

This has been solved partially by updating residue matching algorithm and we deployed the solved version in our web-server.
We're also preparing to deploy superposition improvement.

[timt51]

Great to hear, excited for future improvements!