Take into account pLDDT into TM-score

[mariaartlle]

Hi, I am currently superimposing protein structures of different species with TM-Align and using a TM-score threshold to filter. However, I have realized that, since I am using AlphaFold structures, the TM-score is heavily influenced by low pLDDT regions that appear as disordered coils. As I am trying to annotate proteins, I would like to still get a global TM-score that somehow penalizes the poor predicted regions, as long as they don't represent a large proportion of the protein.

For instance, I have a candidate protein that has an almost perfect superimposition with the core of the reference protein, but as it also has a disordered terminal region, the global TM-score drops and it does not overcome the threshold I defined to filter. Is there a correction measure that takes into account the pLDDT regions when computing the TM-score?

[jlspzw]

Dear user,

If you make sure some areas are disordered, you can remove those residues both from the model and native, only keep the core region that you mentioned, then you will get a good score.

The pLDDT score and TM-score are completely different things, pLDDT score is a predicted LDDT score, so we may not integrate it in TM-score calculation. so you may want to use the method I mentioned to re-calculate the scores.

Best
IT Team

[SandraJKwong]

Challenges in accurately assessing protein structure using TM scores when low pLDDT regions,basket random which are often considered disordered, lead to false positives in the study.

[throughyuletide]

bitlife
Hi, I am currently superimposing protein structures of different species with TM-Align and using a TM-score threshold to filter. However, I have realized that, since I am using AlphaFold structures, the TM-score is heavily influenced by low pLDDT regions that appear as disordered coils. As I am trying to annotate proteins, I would like to still get a global TM-score that somehow penalizes the poor predicted regions, as long as they don't represent a large proportion of the protein.

For instance, I have a candidate protein that has an almost perfect superimposition with the core of the reference protein, but as it also has a disordered terminal region, the global TM-score drops and it does not overcome the threshold I defined to filter. Is there a correction measure that takes into account the pLDDT regions when computing the TM-score?

Mask Low-pLDDT Regions: Remove or mask residues with low pLDDT scores (e.g., below 70) before running TM-align to focus on high-confidence regions only.
Custom Weighted TM-score: Post-process TM-align results by calculating a pLDDT-weighted TM-score where high-confidence residues contribute more to the score.
Try Alternative Alignment Tools: Use alignment tools like Dali or CEalign in PyMOL, which may handle disordered regions differently and provide metrics that prioritize core similarities.