Investigating bias in semantic similarity measures for analysis of protein interactions