Perform so called computational alanine scanning and prediction of hot spot residues.

Although you must refer to the recent literature for the currently best methods and their performances, I can suggest you some to start:

• Rosetta alanine scanning (download Rosetta or try to use it on Robetta server)
• a server (one of several) to predict hot spot residues http://kfc.mitchell-lab.org/upload.php
• FoldX, e.g. mutate all interface residues to alanines one by one and calculate energy differences.

PocketQuery is the successor of ANCHOR. It has pre-calculated delta G (from FastContact, like ANCHOR) and delta delta G (from Rosetta) values for all the PPIs in the PDB. You can also submit your own structure, although this takes longer to process than ANCHOR due to the Rosetta and sequence conservation calculations.

PocketQuery is designed to support the identification of small-molecule inhibitor starting points, so it performs an analysis on all possible clusters of co-located residues and allows you to output the result to various pharmacophore search engines. However, by setting the cluster size to 1 and setting the view to Residue Centric you can explore the PPI interface at a per residue level. Also, the viewer controls allow you to color interface residues based on their properties.