I used AlphaFold to predict the structure of a protein that has not been well studied.
I have a pdb file of the predicted structure.
Now I would like to identify the domains of the protein using this pdb file.
Is there a suitable tool for my purpose?
Have you tried InterPro (https://www.ebi.ac.uk/interpro/) first? You can submit the sequence, get the results, and then analyze your pdb file.
Lorenzo Briganti
Thank you very much for your kind answer!
I had not heard of InterPro and had not tried it. I tried it immediately and got the results. From the results, I found a domain that I didn't know about before.
Actually, I searched DALI ( http://ekhidna2.biocenter.helsinki.fi/dali/ ) for proteins that have similar structures with the predicted structure of my protein by AlphaFold, and found some proteins. I compared them with my protein, and found that my protein have some very similar structures to known proteins, and those structures have domain names based on PDB. Those domains were not identified by InterPro. Is it OK if I say my protein has those domains?
Please correct me if I'm wrong, but it feels to me that AlphaFold can predict structures that were previously difficult to predict, and since protein domains are related to structures, AlphaFold can identify protein domains that were difficult to identify with amino acid sequences and existing tools.
You're welcome, Koji Yoshida !
Actually, you are right. One thing you can always do is search on DALI for similar protein structures. They can say a lot, and is more reliable than doing only the sequence alignment. The sequence alignment is an important first step, if you don't know anything about the protein.
AlphaFold is revolutionary, because it can predict structures with very high accuracy. It has changed Structural Biology, providing good models. However, sometimes it can give a very strange structure, completely different from the reality. That's why people should check wisely the structure, validate and compare to similars whenever possible. For me, it worked very well providing a good model for Molecular Replacement for a protein that I couldn't do experimental phasing. But I had the crystallographic data to back me up.
Best of luck,
Lorenzo
Thank you very much for your lucid explanation, Lorenzo Briganti !
I understand very well.
I’m just using tools of Structural Biology, but your advice was so easy to understand even for me.
Koji Yoshida you're very welcome! Wish you all the best in your work!
Hi ,Koji, Lorenzo, Zahar
Could you help me for this question please, I generated the predicat structure protein by using Alpha fold 2. My question, how to identify the transmembrane domain(TMD) and transmembrane helices(TMs)?
Zahra Liyaghatdar
Thank you for your kind answer!I haven't heard of Swiss model, but I already predicted the structure using AlphaFold.
If I'm wrong, please correct me, but it feels to me that Swiss model is for generating 3D structure of a protein from its amino acid sequence?
Does it give me some information about domains?
Raja Alawar I'm sorry, but I don't know how to identify transmembrane domains.
Zahra Liyaghatdar
I think I'm mistaken. I tried Swiss model, and found some homology domains that are similar to some known proteins. It seems that they are also similar to the structures predicted by AlphaFold, Now I'm more confident that my protein has the domains. Thank you for your help!
Swiss Model is another tool to predict structures (by homology modeling), however AlphaFold is far much better (as you can see from the comparisons between these tools), much more accurate. But for many structures, it is quite similar.
A simple tool might be the Predicted Alignment Error provided with Colab and AlphaFold database predictions. This could be a useful indication, as it will often show "blocks" of higher certainty which may correspond to various domains depending on the target protein. For example, a protein I was working on had a clearly defined region that corresponded to a predicted DNA binding domain. Of course this isn't the most robust or "quantitative" method, but it could be a first indication of something
It depends on how you want to identify domains.
AlphaFold2 uses HMMER(http://hmmer.org/) to search seq DB(Pfam)s, and make alignment files, then based on these alignment files calculate the contact maps for folding. So...
1) aa seq-based domain info:
You can grab it directly from AlphaFold2 after HMMER calculation. However, simply you can put your aa seq into the Pfam website will give a similar result. This kinda method will work only when a similar seq is on the DB. AlphaFold2 does some stupid things when it is given multi-domain protein seq which is not on the Pfam DB.
2) Without having aa seq info, use only the PDB file to predict the domain:
well, this could be a harder question, a) either you write custom python code to calculate the center of mass and define the cutoff value to identify two domains,s or b) use at least two PDB files of the same protein and then align them then calculate the relative deviation of the other. We have a couple of unpublished multi-domain protein structures and compared them with AlphaFold2's performance. it predicted well on domains but not the relative position of each domain.
I think you need to provide more info about why and what is next step or what is the purpose of identifying the domain from AlphaFold2 prediction, if not Pfam domain info?
- Badges
- Science topic
- Similar topics
- Filing