Hi everyone, I am wondering if there is a possibility to predict the protein interactions and finding possible targets for an specific protein of interest using the data from its structure and sequence, not based on the experiments. My protein of interest has not been studied a lot. Now, with the advances in AI tools, is there any possibility to do that?
Thanks a lot for your help in advance.
Not in the way you're imagining I suspect. For binding experiments to be meaningful, you need to have a hypothesis about the types of partners/structures its already likely to bind. I'm not aware of any docking or binding approach that would let you just 'guess' at what out of the possible universe of things it might bind to.
I would start with some non-AI approaches like the STRING database to form some hypotheses about function and possible interactions.
In this instance, AI is probably most helpful in trawling the literature in a similar way to STRING to find mentions of the protein and what its possible roles/targets are.
That said, if the protein is suspected of having a very particular function (e.g. DNA binding), you will be able to infer that by looking at domain conservation with tools like hhpred.
Short answer is I don't think AI is going to magically have solved this (though now Alphafold has been let loose on uniprot etc, maybe there's something new to be unearthed.
Thanks very much Joe for your detailed answer. I totally get your point but I thought there might be some new AI-based techniques, as the STING only suggest based on the research already done before. My protein of interest is an extracellular protein that it is known has some specific functions but no one so far has detected to what it is binding.
Thanks again for your time to answer my question.
No one may have detected what it binds to, but are there hypotheses as to the class of molecule or anything? If it binds to a small molecule versus another protein the approaches you take could be quite different.
Came across this the other day (https://neurosnap.ai/services) which has a lot of different tools which might help, but you'll still need some sort of hypothesis I think.
It's a paid platform (I am not affiliated with it in any way) though, so you might want to look for equivalents elsewhere.
You can look at alphapulldown for predicting protein-protein interactions from structure and seq. I'm not totally convinced myself, but make up your own mind.
Log in to answer this question.
