Question

why would you think that the identity matrix is not good method for calculating a similarity score for protein sequence?

Answer 1

Sequence identity values from multiple sequence alignments are more reliable. Clustal is a program for generating multiple sequence alignments . SIAS is simple and good but you can do the same with ClustalX. Do the alignment or load the alignment in ClustalX and then under trees chose output tree format options and select %identity matrix.