Identifying individual amino acids or domains in Alphafold predictions

Question

[Aaron]

I have generated protein folding predictions with Alphafold but I am having trouble analyzing the predictions - how do people accurately identify individual amino acids or domains in the predictions?

Answer 1

[Mensur Dlakic]

Not sure what your point is in identifying individual amino-acids. Maybe you are thinking about visualizing individual amino-acids that are functionally relevant?

Domain identification is non-trivial, and it requires significant expertise. Depending on your luck, you may be able to get some results by submitting your structures here or here.

Generally speaking, domains may be structural units that are easy to see by eye (like thigh, calf and foot as leg domains), or they may be functional units that are not clearly separable from each other (like individual components of the quadriceps muscle). It may be easier to first identify domains from sequence using one of domain databases such as Pfam.

Here is an example of a protein with 2 domains. Would you know that there are 2 domains in it if they were not uniquely colored? Most people wouldn't.

[Aaron]

Thanks Mensur Dlakic this is really helpful! The protein I’ve modelled is a chimeric sequence so I’m not entirely sure the structure of it will appear in any databases. If there were a feature that could visualize the locations of individual amino acids, I was hoping I could go through enough of the amino acids of the sequence to identify the sequence’s structure, at least for my own reference. Do you know if this might be possible?

[Mensur Dlakic]

If there were a feature that could visualize the locations of individual amino acids, I was hoping I could go through enough of the amino acids of the sequence to identify the sequence’s structure, at least for my own reference.

This does not make any sense to me.

Like I said, if you submit your protein sequence to Pfam, it should be able to give you some kind of domain boundaries even for chimeric sequences. Then you can map those domains onto your structure(s) and see if that looks reasonable.