How to detect divergent domains in AlphaFold models (CDD/InterProscan not working, PyMOL alignment)

Question

[LLG]

Hi all,

I’m trying to reconcile literature-defined domains (I, II, III) with AlphaFold models of homologs. For reference I’m using PDB: 1DLC, where the domains are mapped in the database.

Problem: CDD/Pfam/InterPro only detect the domains in the reference, not in my 3 modeled homologs. When I align the models to 1DLC in PyMOL, the functional domain appears shifted compared to where I expect it based on the literature only.

What I’ve tried so far:

• InterProScan, CDD/SPARCLE on the full-length sequences
• PyMOL 'super' to 1DLC

Questions:

• What tools or workflows would you recommend for detecting divergent or shifted domains in modeled proteins (beyond InterPro/CDD)?
• Any best practices in PyMOL for per-domain alignment/selection, so I can compare homologs domain-by-domain?

Thanks a lot! Any advice or tool suggestions would really help.

Answer 1

[Mensur Dlakic]

There isn't much beyond InterPro and CDD that would automatically detect divergent domains, as those databases already include almost all other domain databases. You can manually build hidden Markov models for divergent domains and try that for detection at a sequence level.

Your divergent domains may be easier to detect at a structural level, by comparing them to known and predicted structures.

• https://www.ebi.ac.uk/msd-srv/ssm/
• https://search.foldseek.com/search

Finally, PyMOL has a very basic structural alignment capabilities and won't be able to globally align proteins with large domain movements. For that I recommend FoldMason.

https://search.foldseek.com/foldmason

[LLG]

Thank you very much, I'll try those!