Improving accuracy in binding site comparison with homology modeling

Conformational changes make the comparison of protein structures difficult. Algorithms that identify small differences in protein structures to identify influences on specificity are particularly affected by molecular flexibility. However, such algorithms typically compare proteins with identical function and varying specificity, causing them to focus on closely related proteins rather than the remote evolutionary homologs sought by most comparison algorithms. This focus inspired us to ask if structure prediction algorithms, which more accurately predict the structures of close evolutionary neighbors, can be used to "remodel" existing structures with the same template, to make the comparison of their binding sites more accurate. Our results, on the enolase superfamily and the tyrosine kinases, reveal that this approach to error reduction is indeed possible, enabling our methods to identify influences on specificity in protein structures that originally could not be compared.