One possible outcome is: (a) if the two proteins bind similar ligands and (b) you know which part of the xray structure of the solved protein is binding the ligand and (c) you can use this information to tweak the alignment and determine that there is higher identity/similarity in the ligand binding region, that would increase your confidence in the prediction of the ligand binding site relative to a case where these factors weren't present.  The local sequence identity in the binding site could be more important than the overall sequence identity.  Some models may be accurate enough for purposes such as identifying domain junctions, or identifying residues that are likely involved in the function of the protein (ie possible ligand binding residues), but not accurate enough for detailed inhibitor design.  So whether you can draw "reliable" conclusions depends on the type of conclusions you are trying to draw.

Another possible outcome is that the model is completely wrong. At low identity levels it would be very important to test any model against experimental data to make sure it isn't junk.  Also make sure your alignment places gaps in sensible locations taking into account the structure of your template.

As a real world example, the lowest identity useful model I've ever made for detailed inhibitor design was a kinase where the sequence identities in the kinase domain were in the very low 20s.  The active site was more conserved though, and since kinase binding sites are very well studied there was additional information to help ensure the alignments in the ligand site were correct.  Even in that case wasn't clear when we made the model if it was going to be useful or not; it was only during the course of inhibitor optimization that we gained more confidence in its accuracy.

Cheers,
Eric



On May 22, 2014, at 3:48 PM, Theresa Hsu wrote:

> Dear all
> 
> I am working with a membrane protein without known structure. The closest protein in PDB has 10% sequence identity/25% similarity to my protein.
> 
> What is the best method and software to do homology modeling while I try to get the crystal? Is the ligand binding site prediction reliable? There is no available experimental data on this protein except to sugest it is some type of ion transport.
> 
> Thank you.
> 
> Theresa

--
Eric Bennett, [log in to unmask]

Always try to associate yourself with and learn as much as you can from those who know more than you do, who do better than you, who see more clearly than you.
- Dwight Eisenhower