Residue-residue contact vs protein-protein interactions

Question

[Anh]

Hi everyone, I'm concern about the problem of predicting binary protein-protein interaction. Two proteins are defined to interact if they form a complex executing a function transiently and stably, e.g. a protein-ligand binds to a protein-receptor, homo-oligomers protein. But I notice that given two imaginary protein sequences folding in the imaginary space will result in two imaginary protein structures. I use the word 'imaginary' here for my thought experiment. Two structures would fluctuate dynamically and atomically, some atoms would attract others, some would repel. And expand to amino acid molecules in protein structures, some amino acids would attract other amino acids, some would repel. These two structures will have some complementary domains, some attracted domains, some repel domains. So can we from residue-residue contact (in this case is the distances in space between amino acids - contact map) to say this protein interacts with this protein or not (from my first definition). Thank you for any of your contributions and arguments.

Answer 1

[Mensur Dlakic]

Contact maps in most cases refer to intra-residue contacts. Still, one can study co-evolution of residues between two different molecules, and that is a legitimate way of predicting interactions. When we talk about distances between residues, usually it is distances that persist over time, and again usually because the protein is folded so its intra-residue distances do not fluctuate.

Two random proteins in cell are like two random strangers in New York City. People walk by each other all the time without making meaningful interactions - and I don't count occasional elbowing as meaningful! Even if they are kind and give each other a smile or a high-five, that would not be much of an interaction (except that we may give it special meaning in our minds). The point is that in an ultra-crowded cellular environment, which is packed with proteins, there will be all kinds of transient bumps and scrapes, but most proteins don't attract or repel each other. They are simply indifferent to each other.

[Anh]

I want to make sure that I got the idea right, sorry for my eager curiosity.

I think the residue-residue contact/distance profile in the interface of any two protein structure pair placed in the same place, that profile will reflect that whether two proteins interact or not (in this case interact here refer to the meaningful interaction that will lead to biological function). If two proteins interact, I think there is a subset of residues of this protein will be close enough to a subset of residues of that protein. In contrast, if two proteins do not interact, there is not any subset of residues of this protein will be close to subset of the other. In some cases, whether exist two proteins that do form any meaningful interact but their residues stay close together in long enough time? I suppose that there is not, they can not get in close long enough if they do not form any meaningful interaction.

I hope that I got a right understanding, if it is not, please let me know. Thank you!

[Mensur Dlakic]

Your understanding is correct. Although proteins bump into each other all the time, which means that subsets of their residues get in close contact with each other, those are not considered true interactions because they do not persist over a period of time.

As I said earlier, distance predictions (and thus contact-map predictions) are usually done intra-protein, meaning residues that are part of the same protein. This is based on co-evolution of interacting residues, where pairs of residues that that are folded in such a way to interact tend to mutate in correlated fashion. It turns out that residues that are part of different proteins that interact also co-evolve (mutate in correlated fashion), which is what they are exploiting in a paper you referenced.

Contact maps derived from intra-residue distances are used as distance restraints for protein folding (and other applications), which is a different application than inter-residue distances in protein interactions, but it is based on the same general principle. Inter-residue distances can be used to restrain the search space for docking protein partners, which not only speeds up the process but also makes it more accurate.

Answer 2

[Anh]

Thank you for your thought-provoking answer. I have another contemplations, from the paper Sequence-based prediction of protein-protein interactions: a structure-aware interpretable deep learning model, I would quote the definition of two proteins interact physically

Our approach encodes the intuition that for two proteins to physically interact, a subset of amino acids from each protein should be in contact with the other.

1. Are these physical interactions the same as interactions in my very first definition (e.g. receptor and ligand, homo and hetero-oligomers of proteins, etc).

2. And if we just consider physical distance or atomical distance between residues of two protein structures in 3D space, can we say they interact, I know some of them are really meaningful interaction, when say meaningful I mean they interact to perform normal biological functions.

3. In the case of binary protein-protein prediction which I consider in the beginning post, probably interactions here refer to meaningful interactions, not physical interactions atomically, but atomical interactions or atomical contacts are subsets of meaningful interactions and necessary to form meaningful interactions. In other words, to form the meaningful interactions between two proteins, they must come to close contact, some of their amino acid residues must be in close contact atomically. What do you think about this?

If I understand anything incorrect, please let me know and correct me. Thank you!