Predicting Protein-Protein Binding

Question

[Ying W]

I have a protein of interest that I know binds to ER-alpha. I have made deletion mutants to narrow down the domain important for ER-alpha binding to about 300 amino acids. There is low/no conservation in this region.

Are there in-silico methods that I can use to predict binding when I know around where the binding site is (300 aa) and also I know the protein that it binds to (ER-alpha is well characterized so I know certain domains where my protein can't bind)

Answer 1

[Spitshine]

The ELM database makes a couple of suggestions for ER-alpha that, depending on your protein, might be useful to your cause.

Answer 2

[Lyco]

To be honest, if you are serious about tackling this problem through bioinformatics, you should get in contact with an expert. Sure, there are tools claiming to address this problem, but none of them is particularly good at that.

There are three principal possibilities here:

• you have an interaction domain
• you have an interaction motif
• this is a 'one-off' interaction -

If there is really 'no conservation' in this region, this would favor the third option. However, some domains and motifs are so poorly conserved that they are hard to spot. I assume that you have already tried Pfam or Interpro on you protein to check for domains. As 'spitshine' suggested, you might also check ELM for a motif (although this rarely works). As far as I know, there are no known ER-alpha interaction domains or motifs.

Assuming that alll of the above fails, I would do the following: First, get a collection of reliable orthologs of your protein from reasonable taxonomic range (wherever you find decent orthologs, it might be advisable to restrict the search to species that are assumed to have ER-alpha). Next, create a multiple alignment of the sequences with a good program (MAFFT, MUSCLE, PROBCONS). Next, use an alignment viewer/editor to look at the alignment and see if there still is 'no consevation' in the region of interest. Set the alignment viewer/editor to a mode where highlighting is based on conservation rather than amino acid properties. If you still don't see anything conserved in this region, there is not much hope. If there is, you could try to derive a consesus sequence and scan this, or if you are really serious, try to go for HMM- or profile-based methods.

If you have come to this point and have further problems, post again!

Answer 3

[Niallhaslam]

Anchor is a tool for predicting protein binding regions in proteins:

In the graphs they plot on that website they also show the disordered regions using IUPred.

We're also preparing a tool to do a similar task, the site is still very much in the development stage but human proteins are set up to work: http://bioware.ucd.ie/~compass/cgi-bin/PHP_helper_files/viewMap.php?jobId=a00BpM&server=slimpred&imageNo=1

You can also see the relative local conservation of the protein sequence - to highlight say conserved Y residues in a motif or similar.

As spitshine suggested ELM is a good place to find known instances of existing protein binding motifs in your sequences. MiniMotif miner is a similar tool with a larger number of less well characterised motifs: http://mnm.engr.uconn.edu/MNM/SMSSearchServlet

Answer 4

[Dataminer]

You can try docking tools like Autodock and Hex.

Answer 5

[Hranjeev]

I was just reading about this docking tool today - GPU.proton.dock. This improved tool seem to be promising and is a possible solution for protein-protein docking in silico.

They also claim to be able to do in silico mutagenesis for the docking experiments something as an advantage to see how significant is the mutagenesis to your protein docking.

They claim to be fast but I still needed to wait on their 'tutorial' walkthrough to calculate and load for at least 10 minutes now and the results have not yet appeared.

Will update once I get to try more proteins with the webserver.