In general, it is rather tricky to get to the structure directly from sequence, unless you know proteins with known structure that show high sequence similarity to your query. Predicting structure using sequence similarity and known structures is known as homology modeling (you might want to check list of structure prediction tools in wikipedia).
Have a look at HOMPPI server to predict interface residues from residues. There is another version from the same group that accepts two proteins (and their sequence) and give you the potential interaction side. You can also check iFrag for the latter purpose.
It is a good idea to keep in mind the expected false positive rate of these predictions (see the publications these tools are introduced for estimated rates).